Publications
Publications resulting from research conducted using Delta appear here. Check back to see how the list of exciting discoveries made using Delta grows.
If you have a publication that should be listed here and isn’t, please share your success with us!
5005740
XHSH9DGT
1
nature
50
default
1
1693
https://delta.ncsa.illinois.edu/wp-content/plugins/zotpress/
%7B%22status%22%3A%22success%22%2C%22updateneeded%22%3Afalse%2C%22instance%22%3Afalse%2C%22meta%22%3A%7B%22request_last%22%3A250%2C%22request_next%22%3A50%2C%22used_cache%22%3Atrue%7D%2C%22data%22%3A%5B%7B%22key%22%3A%228CSHZC3C%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Rapp%20et%20al.%22%2C%22parsedDate%22%3A%222025-03-24%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%3Cdiv%20class%3D%5C%22csl-bib-body%5C%22%20style%3D%5C%22line-height%3A%202%3B%20%5C%22%3E%5Cn%20%20%3Cdiv%20class%3D%5C%22csl-entry%5C%22%20style%3D%5C%22clear%3A%20left%3B%20%5C%22%3E%5Cn%20%20%20%20%3Cdiv%20class%3D%5C%22csl-left-margin%5C%22%20style%3D%5C%22float%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%5C%22%3E1.%3C%5C%2Fdiv%3E%3Cdiv%20class%3D%5C%22csl-right-inline%5C%22%20style%3D%5C%22margin%3A%200%20.4em%200%201.5em%3B%5C%22%3ERapp%2C%20J.%20%3Ci%3Eet%20al.%3C%5C%2Fi%3E%20Design%20of%20Tough%203D%20Printable%20Elastomers%20with%20Human-in-the-Loop%20Reinforcement%20Learning.%20Preprint%20at%20%3Ca%20class%3D%27zp-DOIURL%27%20href%3D%27https%3A%5C%2F%5C%2Fdoi.org%5C%2F10.26434%5C%2Fchemrxiv-2025-w1563%27%3Ehttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.26434%5C%2Fchemrxiv-2025-w1563%3C%5C%2Fa%3E%20%282025%29.%3C%5C%2Fdiv%3E%5Cn%20%20%3C%5C%2Fdiv%3E%5Cn%3C%5C%2Fdiv%3E%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22Design%20of%20Tough%203D%20Printable%20Elastomers%20with%20Human-in-the-Loop%20Reinforcement%20Learning%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Johann%22%2C%22lastName%22%3A%22Rapp%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Dylan%22%2C%22lastName%22%3A%22Anstine%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Filipp%22%2C%22lastName%22%3A%22Gusev%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Filipp%22%2C%22lastName%22%3A%22Nikitin%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Kelly%22%2C%22lastName%22%3A%22Yun%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Mia%22%2C%22lastName%22%3A%22Borden%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Vittal%22%2C%22lastName%22%3A%22Bhat%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Olexandr%22%2C%22lastName%22%3A%22Isayev%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Frank%22%2C%22lastName%22%3A%22Leibfarth%22%7D%5D%2C%22abstractNote%22%3A%22The%20development%20of%20high-performance%20elastomers%20for%20additive%20manufacturing%20requires%20overcoming%20complex%20property%20trade-offs%20that%20challenge%20conventional%20material%20discovery%20pipelines.%20Here%2C%20a%20human-in-the-loop%20reinforcement%20learning%20%28RL%29%20approach%20is%20used%20to%20discover%20exceptional%20polyurethane%20elastomers%20that%20overcome%20pervasive%20stress%5Cu2013strain%20property%20tradeoffs.%20Starting%20with%20a%20diverse%20training%20set%20of%2092%20formulations%2C%20a%20coupled%20multi-component%20reward%20system%20was%20identified%20that%20guides%20RL%20agents%20toward%20materials%20with%20both%20high%20strength%20and%20extensibility.%20Through%20three%20rounds%20of%20iterative%20optimization%20combining%20RL%20predictions%20with%20human%20chemical%20intuition%2C%20we%20identified%20elastomers%20with%20more%20than%20double%20the%20average%20toughness%20compared%20to%20the%20initial%20training%20set.%20The%20final%20exploitation%20round%2C%20aided%20by%20solubility%20prescreening%2C%20predicted%20twelve%20materials%20exhibiting%20both%20high%20strength%20%28%3E10%20MPa%29%20and%20high%20engineering%20strain%20%28%3E200%25%29.%20Analysis%20of%20the%20high%20performing%20materials%20revealed%20structure%5Cu2013property%20insights%2C%20including%20the%20benefits%20of%20high%20molar%20mass%20urethane%20oligomers%2C%20a%20high%20density%20of%20urethane%20functional%20groups%2C%20and%20incorporation%20of%20rigid%20low%20molecular%20weight%20diols%20and%20unsymmetric%20diisocyanates.%20These%20findings%20demonstrate%20that%20machine-guided%2C%20human-augmented%20design%20is%20a%20powerful%20strategy%20for%20accelerating%20polymer%20discovery%20in%20applications%20where%20data%20is%20scarce%20and%20expensive%20to%20acquire%2C%20with%20broad%20applicability%20to%20multi-objective%20materials%20optimization.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222025-03-24%22%2C%22DOI%22%3A%2210.26434%5C%2Fchemrxiv-2025-w1563%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Fchemrxiv.org%5C%2Fengage%5C%2Fchemrxiv%5C%2Farticle-details%5C%2F67ddbe1f81d2151a0257f3e6%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222025-03-27T13%3A50%3A09Z%22%7D%7D%2C%7B%22key%22%3A%22FU2UNXI7%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Yang%20et%20al.%22%2C%22parsedDate%22%3A%222025%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%3Cdiv%20class%3D%5C%22csl-bib-body%5C%22%20style%3D%5C%22line-height%3A%202%3B%20%5C%22%3E%5Cn%20%20%3Cdiv%20class%3D%5C%22csl-entry%5C%22%20style%3D%5C%22clear%3A%20left%3B%20%5C%22%3E%5Cn%20%20%20%20%3Cdiv%20class%3D%5C%22csl-left-margin%5C%22%20style%3D%5C%22float%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%5C%22%3E1.%3C%5C%2Fdiv%3E%3Cdiv%20class%3D%5C%22csl-right-inline%5C%22%20style%3D%5C%22margin%3A%200%20.4em%200%201.5em%3B%5C%22%3EYang%2C%20Y.%2C%20Taherian%2C%20H.%2C%20Kalkhorani%2C%20V.%20A.%20%26amp%3B%20Wang%2C%20D.%20Elevating%20Robust%20Multi-Talker%20ASR%20by%20Decoupling%20Speaker%20Separation%20and%20Speech%20Recognition.%20Preprint%20at%20%3Ca%20class%3D%27zp-DOIURL%27%20href%3D%27https%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2503.17886%27%3Ehttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2503.17886%3C%5C%2Fa%3E%20%282025%29.%3C%5C%2Fdiv%3E%5Cn%20%20%3C%5C%2Fdiv%3E%5Cn%3C%5C%2Fdiv%3E%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22Elevating%20Robust%20Multi-Talker%20ASR%20by%20Decoupling%20Speaker%20Separation%20and%20Speech%20Recognition%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Yufeng%22%2C%22lastName%22%3A%22Yang%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Hassan%22%2C%22lastName%22%3A%22Taherian%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Vahid%20Ahmadi%22%2C%22lastName%22%3A%22Kalkhorani%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22DeLiang%22%2C%22lastName%22%3A%22Wang%22%7D%5D%2C%22abstractNote%22%3A%22Despite%20the%20tremendous%20success%20of%20automatic%20speech%20recognition%20%28ASR%29%20with%20the%20introduction%20of%20deep%20learning%2C%20its%20performance%20is%20still%20unsatisfactory%20in%20many%20real-world%20multi-talker%20scenarios.%20Speaker%20separation%20excels%20in%20separating%20individual%20talkers%20but%2C%20as%20a%20frontend%2C%20it%20introduces%20processing%20artifacts%20that%20degrade%20the%20ASR%20backend%20trained%20on%20clean%20speech.%20As%20a%20result%2C%20mainstream%20robust%20ASR%20systems%20train%20the%20backend%20on%20noisy%20speech%20to%20avoid%20processing%20artifacts.%20In%20this%20work%2C%20we%20propose%20to%20decouple%20the%20training%20of%20the%20speaker%20separation%20frontend%20and%20the%20ASR%20backend%2C%20with%20the%20latter%20trained%20on%20clean%20speech%20only.%20Our%20decoupled%20system%20achieves%205.1%25%20word%20error%20rates%20%28WER%29%20on%20the%20Libri2Mix%20dev%5C%2Ftest%20sets%2C%20significantly%20outperforming%20other%20multi-talker%20ASR%20baselines.%20Its%20effectiveness%20is%20also%20demonstrated%20with%20the%20state-of-the-art%207.60%25%5C%2F5.74%25%20WERs%20on%201-ch%20and%206-ch%20SMS-WSJ.%20Furthermore%2C%20on%20recorded%20LibriCSS%2C%20we%20achieve%20the%20speaker-attributed%20WER%20of%202.92%25.%20These%20state-of-the-art%20results%20suggest%20that%20decoupling%20speaker%20separation%20and%20recognition%20is%20an%20effective%20approach%20to%20elevate%20robust%20multi-talker%20ASR.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222025%22%2C%22DOI%22%3A%2210.48550%5C%2FARXIV.2503.17886%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2503.17886%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222025-03-27T13%3A43%3A26Z%22%7D%7D%2C%7B%22key%22%3A%22YK9MICST%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Yan%20and%20Snir%22%2C%22parsedDate%22%3A%222025%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%3Cdiv%20class%3D%5C%22csl-bib-body%5C%22%20style%3D%5C%22line-height%3A%202%3B%20%5C%22%3E%5Cn%20%20%3Cdiv%20class%3D%5C%22csl-entry%5C%22%20style%3D%5C%22clear%3A%20left%3B%20%5C%22%3E%5Cn%20%20%20%20%3Cdiv%20class%3D%5C%22csl-left-margin%5C%22%20style%3D%5C%22float%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%5C%22%3E1.%3C%5C%2Fdiv%3E%3Cdiv%20class%3D%5C%22csl-right-inline%5C%22%20style%3D%5C%22margin%3A%200%20.4em%200%201.5em%3B%5C%22%3EYan%2C%20J.%20%26amp%3B%20Snir%2C%20M.%20Contemplating%20a%20Lightweight%20Communication%20Interface%20for%20Asynchronous%20Many-Task%20Systems.%20Preprint%20at%20%3Ca%20class%3D%27zp-DOIURL%27%20href%3D%27https%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2503.15400%27%3Ehttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2503.15400%3C%5C%2Fa%3E%20%282025%29.%3C%5C%2Fdiv%3E%5Cn%20%20%3C%5C%2Fdiv%3E%5Cn%3C%5C%2Fdiv%3E%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22Contemplating%20a%20Lightweight%20Communication%20Interface%20for%20Asynchronous%20Many-Task%20Systems%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Jiakun%22%2C%22lastName%22%3A%22Yan%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Marc%22%2C%22lastName%22%3A%22Snir%22%7D%5D%2C%22abstractNote%22%3A%22Asynchronous%20Many-Task%20Systems%20%28AMTs%29%20exhibit%20different%20communication%20patterns%20from%20traditional%20High-Performance%20Computing%20%28HPC%29%20applications%2C%20characterized%20by%20asynchrony%2C%20concurrency%2C%20and%20multithreading.%20Existing%20communication%20libraries%20usually%20do%20not%20support%20AMTs%27%20communication%20requirements%20in%20the%20most%20direct%20and%20efficient%20ways.%20The%20Lightweight%20Communication%20Interface%20%28LCI%29%20is%20an%20experimental%20communication%20library%20aiming%20to%20push%20for%20efficient%20communication%20support%20for%20AMTs.%20This%20paper%20presents%20the%20design%20for%20a%20new%20LCI%20C%2B%2B%20interface%20and%20its%20rationale.%20With%20a%20new%20C%2B%2B%20%5C%5Cemph%7Bobjectized%20flexible%20functions%7D%20idiom%2C%20the%20new%20interface%20aims%20for%20the%20following%20features%3A%20%28a%29%20a%20concise%20but%20expressive%20interface%20for%20all%20common%20point-to-point%20communication%20primitives%20and%20completion%20mechanisms%2C%20%28b%29%20a%20fine-grained%20resource%20mapping%20scheme%20for%20library%20interoperation%2C%20multithreaded%20performance%20isolation%2C%20and%20flexibility%20%28c%29%20a%20set%20of%20optional%20parameters%20and%20overridable%20classes%20for%20users%20to%20incrementally%20fine-tune%20the%20runtime%20behavior.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222025%22%2C%22DOI%22%3A%2210.48550%5C%2FARXIV.2503.15400%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2503.15400%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222025-03-25T12%3A52%3A27Z%22%7D%7D%2C%7B%22key%22%3A%22T4CLTMW7%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Huang%20et%20al.%22%2C%22parsedDate%22%3A%222025-03-20%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%3Cdiv%20class%3D%5C%22csl-bib-body%5C%22%20style%3D%5C%22line-height%3A%202%3B%20%5C%22%3E%5Cn%20%20%3Cdiv%20class%3D%5C%22csl-entry%5C%22%20style%3D%5C%22clear%3A%20left%3B%20%5C%22%3E%5Cn%20%20%20%20%3Cdiv%20class%3D%5C%22csl-left-margin%5C%22%20style%3D%5C%22float%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%5C%22%3E1.%3C%5C%2Fdiv%3E%3Cdiv%20class%3D%5C%22csl-right-inline%5C%22%20style%3D%5C%22margin%3A%200%20.4em%200%201.5em%3B%5C%22%3EHuang%2C%20H.-K.%2C%20Park%2C%20S.%2C%20Villa%2C%20U.%2C%20Wang%2C%20L.%20V.%20%26amp%3B%20Anastasio%2C%20M.%20A.%20Gradient-free%20joint%20reconstruction%20of%20initial%20pressure%20distribution%20and%20wave%20speeds%20in%20transcranial%20photoacoustic%20computed%20tomography.%20in%20%3Ci%3EPhotons%20Plus%20Ultrasound%3A%20Imaging%20and%20Sensing%202025%3C%5C%2Fi%3E%20vol.%2013319%2097%26%23x2013%3B103%20%28SPIE%2C%202025%29.%3C%5C%2Fdiv%3E%5Cn%20%20%3C%5C%2Fdiv%3E%5Cn%3C%5C%2Fdiv%3E%22%2C%22data%22%3A%7B%22itemType%22%3A%22conferencePaper%22%2C%22title%22%3A%22Gradient-free%20joint%20reconstruction%20of%20initial%20pressure%20distribution%20and%20wave%20speeds%20in%20transcranial%20photoacoustic%20computed%20tomography%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Hsuan-Kai%22%2C%22lastName%22%3A%22Huang%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Seonyeong%22%2C%22lastName%22%3A%22Park%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Umberto%22%2C%22lastName%22%3A%22Villa%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Lihong%20V.%22%2C%22lastName%22%3A%22Wang%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Mark%20A.%22%2C%22lastName%22%3A%22Anastasio%22%7D%5D%2C%22abstractNote%22%3A%22Transcranial%20photoacoustic%20computed%20tomography%20%28PACT%29%20shows%20promise%20as%20a%20neuroimaging%20modality%20for%20humans%2C%20but%20image%20reconstruction%20is%20challenging%20due%20to%20skull-induced%20aberrations%20in%20the%20measurement%20data.%20Traditional%20model-based%20reconstruction%20methods%20require%20accurate%20knowledge%20of%20the%20skull%5Cu2019s%20acoustic%20and%20elastic%20parameters%20to%20mitigate%20these.%20Joint%20reconstruction%20methods%20have%20been%20proposed%20to%20concurrently%20estimate%20the%20required%20skull%20parameters%20and%20initial%20pressure%20distribution.%20Compared%20to%20conventional%20model-based%20methods%2C%20the%20additional%20gradient%20computation%20for%20the%20skull%20parameters%20imposes%20a%20significantly%20higher%20computational%20burden.%20This%20study%20aims%20to%20develop%20a%20joint%20reconstruction%20method%20that%20does%20not%20require%20the%20gradient%20computation%20with%20respect%20to%20the%20skull%20parameters%2C%20enabling%20efficient%20and%20effective%20aberration%20compensation%20in%20three-dimensional%20transcranial%20PACT.%20This%20method%20assumes%20that%20the%20skull%20geometry%20is%20known%20from%20adjunct%20CT%20data%20and%20is%20described%20as%20a%20reduced-dimensionality%20parameterized%20skull%20model%2C%20with%20the%20skull%20parameters%20to%20be%20optimized%20being%20the%20compressional%20and%20shear%20wave%20speeds%20for%20the%20skull%20plate%20and%20diplo%5Cu00a8e%20layer.%20The%20joint%20objective%20function%20is%20represented%20as%20a%20bi-level%20optimization%20problem%2C%20with%20the%20outer%20minimization%20problem%20over%20wave%20speeds%20subject%20to%20the%20inner%20minimization%20problem%20over%20the%20initial%20pressure%20distribution.%20The%20outer%20problem%20is%20then%20solved%20using%20a%20derivative-free%20optimization%20approach.%20The%20method%20was%20systematically%20evaluated%20using%20realistic%20three-dimensional%20head%20phantoms.%20The%20results%20establish%20that%20the%20proposed%20method%20can%20yield%20significantly%20improved%20image%20quality%20compared%20to%20a%20model-based%20method%20when%20the%20skull%5Cu2019s%20wave%20speeds%20were%20not%20known%20accurately.%20The%20proposed%20method%20offers%20an%20efficient%20approach%20for%20reconstructing%20images%20of%20subjects%20with%20poorly%20characterized%20skull%20properties.%22%2C%22date%22%3A%222025%5C%2F03%5C%2F20%22%2C%22proceedingsTitle%22%3A%22Photons%20Plus%20Ultrasound%3A%20Imaging%20and%20Sensing%202025%22%2C%22conferenceName%22%3A%22Photons%20Plus%20Ultrasound%3A%20Imaging%20and%20Sensing%202025%22%2C%22language%22%3A%22%22%2C%22DOI%22%3A%2210.1117%5C%2F12.3049081%22%2C%22ISBN%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Fwww.spiedigitallibrary.org%5C%2Fconference-proceedings-of-spie%5C%2F13319%5C%2F133190K%5C%2FGradient-free-joint-reconstruction-of-initial-pressure-distribution-and-wave%5C%2F10.1117%5C%2F12.3049081.full%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222025-03-24T18%3A41%3A12Z%22%7D%7D%2C%7B%22key%22%3A%22YHX86CPY%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Pant%20et%20al.%22%2C%22parsedDate%22%3A%222025-03-18%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%3Cdiv%20class%3D%5C%22csl-bib-body%5C%22%20style%3D%5C%22line-height%3A%202%3B%20%5C%22%3E%5Cn%20%20%3Cdiv%20class%3D%5C%22csl-entry%5C%22%20style%3D%5C%22clear%3A%20left%3B%20%5C%22%3E%5Cn%20%20%20%20%3Cdiv%20class%3D%5C%22csl-left-margin%5C%22%20style%3D%5C%22float%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%5C%22%3E1.%3C%5C%2Fdiv%3E%3Cdiv%20class%3D%5C%22csl-right-inline%5C%22%20style%3D%5C%22margin%3A%200%20.4em%200%201.5em%3B%5C%22%3EPant%2C%20S.%20%3Ci%3Eet%20al.%3C%5C%2Fi%3E%20Dissecting%20Large-Scale%20Structural%20Transitions%20in%20Membrane%20Transporters%20Using%20Advanced%20Simulation%20Technologies.%20%3Ci%3EJ.%20Phys.%20Chem.%20B%3C%5C%2Fi%3E%20acs.jpcb.5c00104%20%282025%29%20%3Ca%20class%3D%27zp-DOIURL%27%20href%3D%27http%3A%5C%2F%5C%2Fdoi.org%5C%2F10.1021%5C%2Facs.jpcb.5c00104%27%3Ehttp%3A%5C%2F%5C%2Fdoi.org%5C%2F10.1021%5C%2Facs.jpcb.5c00104%3C%5C%2Fa%3E.%3C%5C%2Fdiv%3E%5Cn%20%20%3C%5C%2Fdiv%3E%5Cn%3C%5C%2Fdiv%3E%22%2C%22data%22%3A%7B%22itemType%22%3A%22journalArticle%22%2C%22title%22%3A%22Dissecting%20Large-Scale%20Structural%20Transitions%20in%20Membrane%20Transporters%20Using%20Advanced%20Simulation%20Technologies%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Shashank%22%2C%22lastName%22%3A%22Pant%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Sepehr%22%2C%22lastName%22%3A%22Dehghani-Ghahnaviyeh%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Noah%22%2C%22lastName%22%3A%22Trebesch%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Ali%22%2C%22lastName%22%3A%22Rasouli%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Tianle%22%2C%22lastName%22%3A%22Chen%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Karan%22%2C%22lastName%22%3A%22Kapoor%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Po-Chao%22%2C%22lastName%22%3A%22Wen%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Emad%22%2C%22lastName%22%3A%22Tajkhorshid%22%7D%5D%2C%22abstractNote%22%3A%22%22%2C%22date%22%3A%222025-03-18%22%2C%22language%22%3A%22en%22%2C%22DOI%22%3A%2210.1021%5C%2Facs.jpcb.5c00104%22%2C%22ISSN%22%3A%221520-6106%2C%201520-5207%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Fpubs.acs.org%5C%2Fdoi%5C%2F10.1021%5C%2Facs.jpcb.5c00104%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222025-03-21T12%3A41%3A39Z%22%7D%7D%2C%7B%22key%22%3A%22X72SL4GL%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Hu%20et%20al.%22%2C%22parsedDate%22%3A%222025%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%3Cdiv%20class%3D%5C%22csl-bib-body%5C%22%20style%3D%5C%22line-height%3A%202%3B%20%5C%22%3E%5Cn%20%20%3Cdiv%20class%3D%5C%22csl-entry%5C%22%20style%3D%5C%22clear%3A%20left%3B%20%5C%22%3E%5Cn%20%20%20%20%3Cdiv%20class%3D%5C%22csl-left-margin%5C%22%20style%3D%5C%22float%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%5C%22%3E1.%3C%5C%2Fdiv%3E%3Cdiv%20class%3D%5C%22csl-right-inline%5C%22%20style%3D%5C%22margin%3A%200%20.4em%200%201.5em%3B%5C%22%3EHu%2C%20Y.%20%3Ci%3Eet%20al.%3C%5C%2Fi%3E%20Empirical%20Privacy%20Variance.%20Preprint%20at%20%3Ca%20class%3D%27zp-DOIURL%27%20href%3D%27https%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2503.12314%27%3Ehttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2503.12314%3C%5C%2Fa%3E%20%282025%29.%3C%5C%2Fdiv%3E%5Cn%20%20%3C%5C%2Fdiv%3E%5Cn%3C%5C%2Fdiv%3E%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22Empirical%20Privacy%20Variance%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Yuzheng%22%2C%22lastName%22%3A%22Hu%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Fan%22%2C%22lastName%22%3A%22Wu%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Ruicheng%22%2C%22lastName%22%3A%22Xian%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Yuhang%22%2C%22lastName%22%3A%22Liu%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Lydia%22%2C%22lastName%22%3A%22Zakynthinou%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Pritish%22%2C%22lastName%22%3A%22Kamath%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Chiyuan%22%2C%22lastName%22%3A%22Zhang%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22David%22%2C%22lastName%22%3A%22Forsyth%22%7D%5D%2C%22abstractNote%22%3A%22We%20propose%20the%20notion%20of%20empirical%20privacy%20variance%20and%20study%20it%20in%20the%20context%20of%20differentially%20private%20fine-tuning%20of%20language%20models.%20Specifically%2C%20we%20show%20that%20models%20calibrated%20to%20the%20same%20%24%28%5C%5Cvarepsilon%2C%20%5Cu03b4%29%24-DP%20guarantee%20using%20DP-SGD%20with%20different%20hyperparameter%20configurations%20can%20exhibit%20significant%20variations%20in%20empirical%20privacy%2C%20which%20we%20quantify%20through%20the%20lens%20of%20memorization.%20We%20investigate%20the%20generality%20of%20this%20phenomenon%20across%20multiple%20dimensions%20and%20discuss%20why%20it%20is%20surprising%20and%20relevant.%20Through%20regression%20analysis%2C%20we%20examine%20how%20individual%20and%20composite%20hyperparameters%20influence%20empirical%20privacy.%20The%20results%20reveal%20a%20no-free-lunch%20trade-off%3A%20existing%20practices%20of%20hyperparameter%20tuning%20in%20DP-SGD%2C%20which%20focus%20on%20optimizing%20utility%20under%20a%20fixed%20privacy%20budget%2C%20often%20come%20at%20the%20expense%20of%20empirical%20privacy.%20To%20address%20this%2C%20we%20propose%20refined%20heuristics%20for%20hyperparameter%20selection%20that%20explicitly%20account%20for%20empirical%20privacy%2C%20showing%20that%20they%20are%20both%20precise%20and%20practically%20useful.%20Finally%2C%20we%20take%20preliminary%20steps%20to%20understand%20empirical%20privacy%20variance.%20We%20propose%20two%20hypotheses%2C%20identify%20limitations%20in%20existing%20techniques%20like%20privacy%20auditing%2C%20and%20outline%20open%20questions%20for%20future%20research.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222025%22%2C%22DOI%22%3A%2210.48550%5C%2FARXIV.2503.12314%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2503.12314%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222025-03-21T12%3A32%3A12Z%22%7D%7D%2C%7B%22key%22%3A%225RBS4I9J%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Chung%20et%20al.%22%2C%22parsedDate%22%3A%222025%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%3Cdiv%20class%3D%5C%22csl-bib-body%5C%22%20style%3D%5C%22line-height%3A%202%3B%20%5C%22%3E%5Cn%20%20%3Cdiv%20class%3D%5C%22csl-entry%5C%22%20style%3D%5C%22clear%3A%20left%3B%20%5C%22%3E%5Cn%20%20%20%20%3Cdiv%20class%3D%5C%22csl-left-margin%5C%22%20style%3D%5C%22float%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%5C%22%3E1.%3C%5C%2Fdiv%3E%3Cdiv%20class%3D%5C%22csl-right-inline%5C%22%20style%3D%5C%22margin%3A%200%20.4em%200%201.5em%3B%5C%22%3EChung%2C%20A.%20K.-W.%2C%20Lam%2C%20K.%20K.-H.%20%26amp%3B%20Yunes%2C%20N.%20Quasinormal%20mode%20frequencies%20and%20gravitational%20perturbations%20of%20spinning%20black%20holes%20in%20modified%20gravity%20through%20METRICS%3A%20The%20dynamical%20Chern-Simons%20gravity%20case.%20Preprint%20at%20%3Ca%20class%3D%27zp-DOIURL%27%20href%3D%27https%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2503.11759%27%3Ehttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2503.11759%3C%5C%2Fa%3E%20%282025%29.%3C%5C%2Fdiv%3E%5Cn%20%20%3C%5C%2Fdiv%3E%5Cn%3C%5C%2Fdiv%3E%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22Quasinormal%20mode%20frequencies%20and%20gravitational%20perturbations%20of%20spinning%20black%20holes%20in%20modified%20gravity%20through%20METRICS%3A%20The%20dynamical%20Chern-Simons%20gravity%20case%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Adrian%20Ka-Wai%22%2C%22lastName%22%3A%22Chung%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Kelvin%20Ka-Ho%22%2C%22lastName%22%3A%22Lam%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Nicolas%22%2C%22lastName%22%3A%22Yunes%22%7D%5D%2C%22abstractNote%22%3A%22We%20present%20the%20first%20precise%20calculations%20of%20the%20gravitational%20quasinormal-mode%20%28QNM%29%20frequencies%20for%20spinning%20black%20holes%20with%20dimensionless%20angular%20momenta%20%24J%5C%2FM%5E2%20%3A%3D%20a%20%5C%5Clesssim%200.75%24%20in%20dynamical%20Chern-Simons%20gravity.%20Using%20the%20%5C%5Ctextit%7BMetric%20pErTuRbations%20wIth%20speCtral%20methodS%7D%20%28METRICS%29%20framework%2C%20we%20compute%20the%20QNM%20frequencies%20of%20both%20axial%20and%20polar%20metric%20perturbations%2C%20focusing%20on%20the%20%24nl%20m%20%3D%20022%24%2C%20%24033%24%2C%20and%20%24032%24%20modes.%20The%20METRICS%20frequencies%20for%20the%20022%20mode%20achieve%20numerical%20uncertainties%20%24%5C%5Clesssim%2010%5E%7B-4%7D%24%20when%20%240%20%5C%5Cleq%20a%20%5C%5Cleq%200.5%24%20and%20%24%5C%5Clesssim%2010%5E%7B-3%7D%24%20for%20%240.5%20%5C%5Cleq%20a%20%5C%5Cleq%200.75%24%2C%20without%20decoupling%20or%20simplifying%20the%20linearized%20field%20equations.%20We%20also%20derive%20optimal%20fitting%20polynomials%20to%20enable%20efficient%20and%20accurate%20evaluations%20of%20the%20leading-order%20frequency%20shifts%20in%20these%20modes.%20The%20METRICS%20frequencies%20and%20fitting%20expressions%20are%20a%20robust%20and%20indispensable%20step%20toward%20enabling%20gravitational-wave%20ringdown%20tests%20of%20dynamical%20Chern-Simons%20gravity.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222025%22%2C%22DOI%22%3A%2210.48550%5C%2FARXIV.2503.11759%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2503.11759%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222025-03-21T12%3A27%3A01Z%22%7D%7D%2C%7B%22key%22%3A%22BJP547HH%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Yan%20et%20al.%22%2C%22parsedDate%22%3A%222025%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%3Cdiv%20class%3D%5C%22csl-bib-body%5C%22%20style%3D%5C%22line-height%3A%202%3B%20%5C%22%3E%5Cn%20%20%3Cdiv%20class%3D%5C%22csl-entry%5C%22%20style%3D%5C%22clear%3A%20left%3B%20%5C%22%3E%5Cn%20%20%20%20%3Cdiv%20class%3D%5C%22csl-left-margin%5C%22%20style%3D%5C%22float%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%5C%22%3E1.%3C%5C%2Fdiv%3E%3Cdiv%20class%3D%5C%22csl-right-inline%5C%22%20style%3D%5C%22margin%3A%200%20.4em%200%201.5em%3B%5C%22%3EYan%2C%20J.%2C%20Kaiser%2C%20H.%20%26amp%3B%20Snir%2C%20M.%20Understanding%20the%20Communication%20Needs%20of%20Asynchronous%20Many-Task%20Systems%20--%20A%20Case%20Study%20of%20HPX%2BLCI.%20Preprint%20at%20%3Ca%20class%3D%27zp-DOIURL%27%20href%3D%27https%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2503.12774%27%3Ehttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2503.12774%3C%5C%2Fa%3E%20%282025%29.%3C%5C%2Fdiv%3E%5Cn%20%20%3C%5C%2Fdiv%3E%5Cn%3C%5C%2Fdiv%3E%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22Understanding%20the%20Communication%20Needs%20of%20Asynchronous%20Many-Task%20Systems%20--%20A%20Case%20Study%20of%20HPX%2BLCI%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Jiakun%22%2C%22lastName%22%3A%22Yan%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Hartmut%22%2C%22lastName%22%3A%22Kaiser%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Marc%22%2C%22lastName%22%3A%22Snir%22%7D%5D%2C%22abstractNote%22%3A%22Asynchronous%20Many-Task%20%28AMT%29%20systems%20offer%20a%20potential%20solution%20for%20efficiently%20programming%20complicated%20scientific%20applications%20on%20extreme-scale%20heterogeneous%20architectures.%20However%2C%20they%20exhibit%20different%20communication%20needs%20from%20traditional%20bulk-synchronous%20parallel%20%28BSP%29%20applications%2C%20posing%20new%20challenges%20for%20underlying%20communication%20libraries.%20This%20work%20systematically%20studies%20the%20communication%20needs%20of%20AMTs%20and%20explores%20how%20communication%20libraries%20can%20be%20structured%20to%20better%20satisfy%20them%20through%20a%20case%20study%20of%20a%20real-world%20AMT%20system%2C%20HPX.%20We%20first%20examine%20its%20communication%20stack%20layout%20and%20formalize%20the%20communication%20abstraction%20that%20underlying%20communication%20libraries%20need%20to%20support.%20We%20then%20analyze%20its%20current%20MPI%20backend%20%28parcelport%29%20and%20identify%20four%20categories%20of%20needs%20that%20are%20not%20typical%20in%20the%20BSP%20model%20and%20are%20not%20well%20covered%20by%20the%20MPI%20standard.%20To%20bridge%20these%20gaps%2C%20we%20design%20from%20the%20native%20network%20layer%20and%20incorporate%20various%20techniques%2C%20including%20one-sided%20communication%2C%20queue-based%20completion%20notification%2C%20explicit%20progressing%2C%20and%20different%20ways%20of%20resource%20contention%20mitigation%2C%20in%20a%20new%20parcelport%20with%20an%20experimental%20communication%20library%2C%20LCI.%20Overall%2C%20the%20resulting%20LCI%20parcelport%20outperforms%20the%20existing%20MPI%20parcelport%20with%20up%20to%2050x%20in%20microbenchmarks%20and%202x%20in%20a%20real-world%20application.%20Using%20it%20as%20a%20testbed%2C%20we%20design%20LCI%20parcelport%20variants%20to%20quantify%20the%20performance%20contributions%20of%20each%20technique.%20This%20work%20combines%20conceptual%20analysis%20and%20experiment%20results%20to%20offer%20a%20practical%20guideline%20for%20the%20future%20development%20of%20communication%20libraries%20and%20AMT%20communication%20layers.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222025%22%2C%22DOI%22%3A%2210.48550%5C%2FARXIV.2503.12774%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2503.12774%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222025-03-21T12%3A23%3A08Z%22%7D%7D%2C%7B%22key%22%3A%22JPGA3ZMT%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Merzky%20et%20al.%22%2C%22parsedDate%22%3A%222025%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%3Cdiv%20class%3D%5C%22csl-bib-body%5C%22%20style%3D%5C%22line-height%3A%202%3B%20%5C%22%3E%5Cn%20%20%3Cdiv%20class%3D%5C%22csl-entry%5C%22%20style%3D%5C%22clear%3A%20left%3B%20%5C%22%3E%5Cn%20%20%20%20%3Cdiv%20class%3D%5C%22csl-left-margin%5C%22%20style%3D%5C%22float%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%5C%22%3E1.%3C%5C%2Fdiv%3E%3Cdiv%20class%3D%5C%22csl-right-inline%5C%22%20style%3D%5C%22margin%3A%200%20.4em%200%201.5em%3B%5C%22%3EMerzky%2C%20A.%20%3Ci%3Eet%20al.%3C%5C%2Fi%3E%20Scalable%20Runtime%20Architecture%20for%20Data-driven%2C%20Hybrid%20HPC%20and%20ML%20Workflow%20Applications.%20Preprint%20at%20%3Ca%20class%3D%27zp-DOIURL%27%20href%3D%27https%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2503.13343%27%3Ehttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2503.13343%3C%5C%2Fa%3E%20%282025%29.%3C%5C%2Fdiv%3E%5Cn%20%20%3C%5C%2Fdiv%3E%5Cn%3C%5C%2Fdiv%3E%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22Scalable%20Runtime%20Architecture%20for%20Data-driven%2C%20Hybrid%20HPC%20and%20ML%20Workflow%20Applications%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Andre%22%2C%22lastName%22%3A%22Merzky%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Mikhail%22%2C%22lastName%22%3A%22Titov%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Matteo%22%2C%22lastName%22%3A%22Turilli%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Ozgur%22%2C%22lastName%22%3A%22Kilic%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Tianle%22%2C%22lastName%22%3A%22Wang%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Shantenu%22%2C%22lastName%22%3A%22Jha%22%7D%5D%2C%22abstractNote%22%3A%22Hybrid%20workflows%20combining%20traditional%20HPC%20and%20novel%20ML%20methodologies%20are%20transforming%20scientific%20computing.%20This%20paper%20presents%20the%20architecture%20and%20implementation%20of%20a%20scalable%20runtime%20system%20that%20extends%20RADICAL-Pilot%20with%20service-based%20execution%20to%20support%20AI-out-HPC%20workflows.%20Our%20runtime%20system%20enables%20distributed%20ML%20capabilities%2C%20efficient%20resource%20management%2C%20and%20seamless%20HPC%5C%2FML%20coupling%20across%20local%20and%20remote%20platforms.%20Preliminary%20experimental%20results%20show%20that%20our%20approach%20manages%20concurrent%20execution%20of%20ML%20models%20across%20local%20and%20remote%20HPC%5C%2Fcloud%20resources%20with%20minimal%20architectural%20overheads.%20This%20lays%20the%20foundation%20for%20prototyping%20three%20representative%20data-driven%20workflow%20applications%20and%20executing%20them%20at%20scale%20on%20leadership-class%20HPC%20platforms.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222025%22%2C%22DOI%22%3A%2210.48550%5C%2FARXIV.2503.13343%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2503.13343%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222025-03-20T14%3A59%3A23Z%22%7D%7D%2C%7B%22key%22%3A%22EGPNTPQA%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Cui%20et%20al.%22%2C%22parsedDate%22%3A%222025%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%3Cdiv%20class%3D%5C%22csl-bib-body%5C%22%20style%3D%5C%22line-height%3A%202%3B%20%5C%22%3E%5Cn%20%20%3Cdiv%20class%3D%5C%22csl-entry%5C%22%20style%3D%5C%22clear%3A%20left%3B%20%5C%22%3E%5Cn%20%20%20%20%3Cdiv%20class%3D%5C%22csl-left-margin%5C%22%20style%3D%5C%22float%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%5C%22%3E1.%3C%5C%2Fdiv%3E%3Cdiv%20class%3D%5C%22csl-right-inline%5C%22%20style%3D%5C%22margin%3A%200%20.4em%200%201.5em%3B%5C%22%3ECui%2C%20S.%20%3Ci%3Eet%20al.%3C%5C%2Fi%3E%20Characterizing%20GPU%20Resilience%20and%20Impact%20on%20AI%5C%2FHPC%20Systems.%20Preprint%20at%20%3Ca%20class%3D%27zp-DOIURL%27%20href%3D%27https%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2503.11901%27%3Ehttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2503.11901%3C%5C%2Fa%3E%20%282025%29.%3C%5C%2Fdiv%3E%5Cn%20%20%3C%5C%2Fdiv%3E%5Cn%3C%5C%2Fdiv%3E%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22Characterizing%20GPU%20Resilience%20and%20Impact%20on%20AI%5C%2FHPC%20Systems%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Shengkun%22%2C%22lastName%22%3A%22Cui%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Archit%22%2C%22lastName%22%3A%22Patke%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Ziheng%22%2C%22lastName%22%3A%22Chen%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Aditya%22%2C%22lastName%22%3A%22Ranjan%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Hung%22%2C%22lastName%22%3A%22Nguyen%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Phuong%22%2C%22lastName%22%3A%22Cao%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Saurabh%22%2C%22lastName%22%3A%22Jha%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Brett%22%2C%22lastName%22%3A%22Bode%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Gregory%22%2C%22lastName%22%3A%22Bauer%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Chandra%22%2C%22lastName%22%3A%22Narayanaswami%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Daby%22%2C%22lastName%22%3A%22Sow%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Catello%22%2C%22lastName%22%3A%22Di%20Martino%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Zbigniew%20T.%22%2C%22lastName%22%3A%22Kalbarczyk%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Ravishankar%20K.%22%2C%22lastName%22%3A%22Iyer%22%7D%5D%2C%22abstractNote%22%3A%22In%20this%20study%2C%20we%20characterize%20GPU%20failures%20in%20Delta%2C%20the%20current%20large-scale%20AI%20system%20with%20over%20600%20petaflops%20of%20peak%20compute%20throughput.%20The%20system%20comprises%20GPU%20and%20non-GPU%20nodes%20with%20modern%20AI%20accelerators%2C%20such%20as%20NVIDIA%20A40%2C%20A100%2C%20and%20H100%20GPUs.%20The%20study%20uses%20two%20and%20a%20half%20years%20of%20data%20on%20GPU%20errors.%20We%20evaluate%20the%20resilience%20of%20GPU%20hardware%20components%20to%20determine%20the%20vulnerability%20of%20different%20GPU%20components%20to%20failure%20and%20their%20impact%20on%20the%20GPU%20and%20node%20availability.%20We%20measure%20the%20key%20propagation%20paths%20in%20GPU%20hardware%2C%20GPU%20interconnect%20%28NVLink%29%2C%20and%20GPU%20memory.%20Finally%2C%20we%20evaluate%20the%20impact%20of%20the%20observed%20GPU%20errors%20on%20user%20jobs.%20Our%20key%20findings%20are%3A%20%28i%29%20Contrary%20to%20common%20beliefs%2C%20GPU%20memory%20is%20over%2030x%20more%20reliable%20than%20GPU%20hardware%20in%20terms%20of%20MTBE%20%28mean%20time%20between%20errors%29.%20%28ii%29%20The%20newly%20introduced%20GSP%20%28GPU%20System%20Processor%29%20is%20the%20most%20vulnerable%20GPU%20hardware%20component.%20%28iii%29%20NVLink%20errors%20did%20not%20always%20lead%20to%20user%20job%20failure%2C%20and%20we%20attribute%20it%20to%20the%20underlying%20error%20detection%20and%20retry%20mechanisms%20employed.%20%28iv%29%20We%20show%20multiple%20examples%20of%20hardware%20errors%20originating%20from%20one%20of%20the%20key%20GPU%20hardware%20components%2C%20leading%20to%20application%20failure.%20%28v%29%20We%20project%20the%20impact%20of%20GPU%20node%20availability%20on%20larger%20scales%20with%20emulation%20and%20find%20that%20significant%20overprovisioning%20between%205-20%25%20would%20be%20necessary%20to%20handle%20GPU%20failures.%20If%20GPU%20availability%20were%20improved%20to%2099.9%25%2C%20the%20overprovisioning%20would%20be%20reduced%20by%204x.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222025%22%2C%22DOI%22%3A%2210.48550%5C%2FARXIV.2503.11901%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2503.11901%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222025-03-20T14%3A45%3A43Z%22%7D%7D%2C%7B%22key%22%3A%22KSHL8V4Y%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22You%20and%20Guo%22%2C%22parsedDate%22%3A%222025%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%3Cdiv%20class%3D%5C%22csl-bib-body%5C%22%20style%3D%5C%22line-height%3A%202%3B%20%5C%22%3E%5Cn%20%20%3Cdiv%20class%3D%5C%22csl-entry%5C%22%20style%3D%5C%22clear%3A%20left%3B%20%5C%22%3E%5Cn%20%20%20%20%3Cdiv%20class%3D%5C%22csl-left-margin%5C%22%20style%3D%5C%22float%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%5C%22%3E1.%3C%5C%2Fdiv%3E%3Cdiv%20class%3D%5C%22csl-right-inline%5C%22%20style%3D%5C%22margin%3A%200%20.4em%200%201.5em%3B%5C%22%3EYou%2C%20Z.%20%26amp%3B%20Guo%2C%20Y.%20PlainQAFact%3A%20Automatic%20Factuality%20Evaluation%20Metric%20for%20Biomedical%20Plain%20Language%20Summaries%20Generation.%20Preprint%20at%20%3Ca%20class%3D%27zp-DOIURL%27%20href%3D%27https%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2503.08890%27%3Ehttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2503.08890%3C%5C%2Fa%3E%20%282025%29.%3C%5C%2Fdiv%3E%5Cn%20%20%3C%5C%2Fdiv%3E%5Cn%3C%5C%2Fdiv%3E%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22PlainQAFact%3A%20Automatic%20Factuality%20Evaluation%20Metric%20for%20Biomedical%20Plain%20Language%20Summaries%20Generation%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Zhiwen%22%2C%22lastName%22%3A%22You%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Yue%22%2C%22lastName%22%3A%22Guo%22%7D%5D%2C%22abstractNote%22%3A%22Hallucinated%20outputs%20from%20language%20models%20pose%20risks%20in%20the%20medical%20domain%2C%20especially%20for%20lay%20audiences%20making%20health-related%20decisions.%20Existing%20factuality%20evaluation%20methods%2C%20such%20as%20entailment-%20and%20question-answering-based%20%28QA%29%2C%20struggle%20with%20plain%20language%20summary%20%28PLS%29%20generation%20due%20to%20elaborative%20explanation%20phenomenon%2C%20which%20introduces%20external%20content%20%28e.g.%2C%20definitions%2C%20background%2C%20examples%29%20absent%20from%20the%20source%20document%20to%20enhance%20comprehension.%20To%20address%20this%2C%20we%20introduce%20PlainQAFact%2C%20a%20framework%20trained%20on%20a%20fine-grained%2C%20human-annotated%20dataset%20PlainFact%2C%20to%20evaluate%20the%20factuality%20of%20both%20source-simplified%20and%20elaboratively%20explained%20sentences.%20PlainQAFact%20first%20classifies%20factuality%20type%20and%20then%20assesses%20factuality%20using%20a%20retrieval-augmented%20QA-based%20scoring%20method.%20Our%20approach%20is%20lightweight%20and%20computationally%20efficient.%20Empirical%20results%20show%20that%20existing%20factuality%20metrics%20fail%20to%20effectively%20evaluate%20factuality%20in%20PLS%2C%20especially%20for%20elaborative%20explanations%2C%20whereas%20PlainQAFact%20achieves%20state-of-the-art%20performance.%20We%20further%20analyze%20its%20effectiveness%20across%20external%20knowledge%20sources%2C%20answer%20extraction%20strategies%2C%20overlap%20measures%2C%20and%20document%20granularity%20levels%2C%20refining%20its%20overall%20factuality%20assessment.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222025%22%2C%22DOI%22%3A%2210.48550%5C%2FARXIV.2503.08890%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2503.08890%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222025-03-18T14%3A17%3A22Z%22%7D%7D%2C%7B%22key%22%3A%22RBRUM3W9%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Balaji%20et%20al.%22%2C%22parsedDate%22%3A%222025%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%3Cdiv%20class%3D%5C%22csl-bib-body%5C%22%20style%3D%5C%22line-height%3A%202%3B%20%5C%22%3E%5Cn%20%20%3Cdiv%20class%3D%5C%22csl-entry%5C%22%20style%3D%5C%22clear%3A%20left%3B%20%5C%22%3E%5Cn%20%20%20%20%3Cdiv%20class%3D%5C%22csl-left-margin%5C%22%20style%3D%5C%22float%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%5C%22%3E1.%3C%5C%2Fdiv%3E%3Cdiv%20class%3D%5C%22csl-right-inline%5C%22%20style%3D%5C%22margin%3A%200%20.4em%200%201.5em%3B%5C%22%3EBalaji%2C%20P.%20%3Ci%3Eet%20al.%3C%5C%2Fi%3E%20Quantum%20Circuits%20for%20SU%283%29%20Lattice%20Gauge%20Theory.%20Preprint%20at%20%3Ca%20class%3D%27zp-DOIURL%27%20href%3D%27https%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2503.08866%27%3Ehttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2503.08866%3C%5C%2Fa%3E%20%282025%29.%3C%5C%2Fdiv%3E%5Cn%20%20%3C%5C%2Fdiv%3E%5Cn%3C%5C%2Fdiv%3E%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22Quantum%20Circuits%20for%20SU%283%29%20Lattice%20Gauge%20Theory%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Praveen%22%2C%22lastName%22%3A%22Balaji%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Cian%5Cu00e1n%22%2C%22lastName%22%3A%22Conefrey-Shinozaki%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Patrick%22%2C%22lastName%22%3A%22Draper%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Jason%20K.%22%2C%22lastName%22%3A%22Elhaderi%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Drishti%22%2C%22lastName%22%3A%22Gupta%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Luis%22%2C%22lastName%22%3A%22Hidalgo%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Andrew%22%2C%22lastName%22%3A%22Lytle%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Enrico%22%2C%22lastName%22%3A%22Rinaldi%22%7D%5D%2C%22abstractNote%22%3A%22Lattice%20gauge%20theories%20in%20varying%20dimensions%2C%20lattice%20volumes%2C%20and%20truncations%20offer%20a%20rich%20family%20of%20targets%20for%20Hamiltonian%20simulation%20on%20quantum%20devices.%20In%20return%2C%20formulating%20quantum%20simulations%20can%20provide%20new%20ways%20of%20thinking%20about%20the%20quantum%20structure%20of%20gauge%20theories.%20In%20this%20work%2C%20we%20consider%20pure%20%24SU%283%29%24%20gauge%20theory%20in%20two%20and%20three%20spatial%20dimensions%20in%20a%20streamlined%20version%20of%20the%20electric%20basis.%20We%20use%20a%20formulation%20of%20the%20theory%20that%20balances%20locality%20of%20the%20Hamiltonian%20and%20size%20of%20the%20gauge-invariant%20state%20space%2C%20and%20we%20classically%20pre-compute%20dictionaries%20of%20plaquette%20operator%20matrix%20elements%20for%20use%20in%20circuit%20construction.%20We%20build%20circuits%20for%20simulating%20time%20evolution%20on%20arbitrary%20lattice%20volumes%2C%20spanning%20circuits%20suitable%20for%20NISQ%20era%20hardware%20to%20future%20fault-tolerant%20devices.%20Relative%20to%20spin%20models%2C%20time%20evolution%20in%20lattice%20gauge%20theories%20involves%20more%20complex%20local%20unitaries%2C%20and%20the%20Hilbert%20space%20of%20all%20quantum%20registers%20may%20have%20large%20unphysical%20subspaces.%20Based%20on%20these%20features%2C%20we%20develop%20general%2C%20volume-scalable%20tools%20for%20optimizing%20circuit%20depth%2C%20including%20pruning%20and%20fusion%20algorithms%20for%20collections%20of%20large%20multi-controlled%20unitaries.%20We%20describe%20scalings%20of%20quantum%20resources%20needed%20to%20simulate%20larger%20circuits%20and%20some%20directions%20for%20future%20algorithmic%20development.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222025%22%2C%22DOI%22%3A%2210.48550%5C%2FARXIV.2503.08866%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2503.08866%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222025-03-18T14%3A03%3A27Z%22%7D%7D%2C%7B%22key%22%3A%22RQK4ZJGX%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Sasidharan%20et%20al.%22%2C%22parsedDate%22%3A%222025%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%3Cdiv%20class%3D%5C%22csl-bib-body%5C%22%20style%3D%5C%22line-height%3A%202%3B%20%5C%22%3E%5Cn%20%20%3Cdiv%20class%3D%5C%22csl-entry%5C%22%20style%3D%5C%22clear%3A%20left%3B%20%5C%22%3E%5Cn%20%20%20%20%3Cdiv%20class%3D%5C%22csl-left-margin%5C%22%20style%3D%5C%22float%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%5C%22%3E1.%3C%5C%2Fdiv%3E%3Cdiv%20class%3D%5C%22csl-right-inline%5C%22%20style%3D%5C%22margin%3A%200%20.4em%200%201.5em%3B%5C%22%3ESasidharan%2C%20A.%2C%20Xian-He%2C%20Lofstead%2C%20J.%20%26amp%3B%20Klasky%2C%20S.%20Performance%20Models%20for%20a%20Two-tiered%20Storage%20System.%20Preprint%20at%20%3Ca%20class%3D%27zp-DOIURL%27%20href%3D%27https%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2503.08966%27%3Ehttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2503.08966%3C%5C%2Fa%3E%20%282025%29.%3C%5C%2Fdiv%3E%5Cn%20%20%3C%5C%2Fdiv%3E%5Cn%3C%5C%2Fdiv%3E%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22Performance%20Models%20for%20a%20Two-tiered%20Storage%20System%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Aparna%22%2C%22lastName%22%3A%22Sasidharan%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22name%22%3A%22Xian-He%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Jay%22%2C%22lastName%22%3A%22Lofstead%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Scott%22%2C%22lastName%22%3A%22Klasky%22%7D%5D%2C%22abstractNote%22%3A%22This%20work%20describes%20the%20design%2C%20implementation%20and%20performance%20analysis%20of%20a%20distributed%20two-tiered%20storage%20software.%20The%20first%20tier%20functions%20as%20a%20distributed%20software%20cache%20implemented%20using%20solid-state%20devices~%28NVMes%29%20and%20the%20second%20tier%20consists%20of%20multiple%20hard%20disks~%28HDDs%29.%20We%20describe%20an%20online%20learning%20algorithm%20that%20manages%20data%20movement%20between%20the%20tiers.%20The%20software%20is%20hybrid%2C%20i.e.%20both%20distributed%20and%20multi-threaded.%20The%20end-to-end%20performance%20model%20of%20the%20two-tier%20system%20was%20developed%20using%20queuing%20networks%20and%20behavioral%20models%20of%20storage%20devices.%20We%20identified%20significant%20parameters%20that%20affect%20the%20performance%20of%20storage%20devices%20and%20created%20behavioral%20models%20for%20each%20device.%20The%20performance%20of%20the%20software%20was%20evaluated%20on%20a%20many-core%20cluster%20using%20non-trivial%20read%5C%2Fwrite%20workloads.%20The%20paper%20provides%20examples%20to%20illustrate%20the%20use%20of%20these%20models.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222025%22%2C%22DOI%22%3A%2210.48550%5C%2FARXIV.2503.08966%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2503.08966%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222025-03-18T13%3A58%3A42Z%22%7D%7D%2C%7B%22key%22%3A%22ZKVES438%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Prather%22%2C%22parsedDate%22%3A%222024%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%3Cdiv%20class%3D%5C%22csl-bib-body%5C%22%20style%3D%5C%22line-height%3A%202%3B%20%5C%22%3E%5Cn%20%20%3Cdiv%20class%3D%5C%22csl-entry%5C%22%20style%3D%5C%22clear%3A%20left%3B%20%5C%22%3E%5Cn%20%20%20%20%3Cdiv%20class%3D%5C%22csl-left-margin%5C%22%20style%3D%5C%22float%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%5C%22%3E1.%3C%5C%2Fdiv%3E%3Cdiv%20class%3D%5C%22csl-right-inline%5C%22%20style%3D%5C%22margin%3A%200%20.4em%200%201.5em%3B%5C%22%3EPrather%2C%20B.%20S.%20KHARMA%3A%20Flexible%2C%20Portable%20Performance%20for%20GRMHD.%20Preprint%20at%20%3Ca%20class%3D%27zp-DOIURL%27%20href%3D%27https%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2408.01361%27%3Ehttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2408.01361%3C%5C%2Fa%3E%20%282024%29.%3C%5C%2Fdiv%3E%5Cn%20%20%3C%5C%2Fdiv%3E%5Cn%3C%5C%2Fdiv%3E%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22KHARMA%3A%20Flexible%2C%20Portable%20Performance%20for%20GRMHD%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Ben%20S.%22%2C%22lastName%22%3A%22Prather%22%7D%5D%2C%22abstractNote%22%3A%22KHARMA%20%28an%20acronym%20for%20%5C%22Kokkos-based%20High-Accuracy%20Relativistic%20Magnetohydrodynamics%20with%20Adaptive%20mesh%20refinement%5C%22%29%20is%20a%20new%20open-source%20code%20for%20conducting%20general-relativistic%20magnetohydrodynamic%20simulations%20in%20stationary%20spacetimes%2C%20primarily%20of%20accretion%20systems.%20It%20implements%20among%20other%20options%20the%20High-Accuracy%20Relativistic%20Magnetohydrodynamics%20%28HARM%29%20scheme%2C%20but%20is%20written%20from%20scratch%20in%20C%2B%2B%20with%20the%20Kokkos%20programming%20model%20in%20order%20to%20run%20efficiently%20on%20both%20CPUs%20and%20GPUs.%20In%20addition%20to%20being%20fast%2C%20KHARMA%20is%20written%20to%20be%20readable%2C%20modular%2C%20and%20extensible%2C%20separating%20functionality%20into%20%5C%22packages%2C%5C%22%20representing%2C%20e.g.%2C%20algorithmic%20components%20or%20physics%20extensions.%20Components%20of%20the%20core%20ideal%20GRMHD%20scheme%20can%20be%20swapped%20at%20runtime%2C%20and%20additional%20packages%20are%20included%20to%20simulate%20electron%20temperature%20evolution%2C%20viscous%20hydrodynamics%2C%20and%20for%20designing%20chained%20multi-scale%20%5C%22bridged%5C%22%20simulations.%20This%20chapter%20presents%20the%20computational%20environment%20and%20requirements%20for%20KHARMA%2C%20features%20and%20design%20which%20meet%20these%20requirements%2C%20and%20finally%2C%20validation%20and%20performance%20data.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222024%22%2C%22DOI%22%3A%2210.48550%5C%2FARXIV.2408.01361%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2408.01361%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222025-03-17T15%3A41%3A59Z%22%7D%7D%2C%7B%22key%22%3A%22FDBTDAWV%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Arora%20et%20al.%22%2C%22parsedDate%22%3A%222025%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%3Cdiv%20class%3D%5C%22csl-bib-body%5C%22%20style%3D%5C%22line-height%3A%202%3B%20%5C%22%3E%5Cn%20%20%3Cdiv%20class%3D%5C%22csl-entry%5C%22%20style%3D%5C%22clear%3A%20left%3B%20%5C%22%3E%5Cn%20%20%20%20%3Cdiv%20class%3D%5C%22csl-left-margin%5C%22%20style%3D%5C%22float%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%5C%22%3E1.%3C%5C%2Fdiv%3E%3Cdiv%20class%3D%5C%22csl-right-inline%5C%22%20style%3D%5C%22margin%3A%200%20.4em%200%201.5em%3B%5C%22%3EArora%2C%20S.%20%3Ci%3Eet%20al.%3C%5C%2Fi%3E%20ESPnet-SDS%3A%20Unified%20Toolkit%20and%20Demo%20for%20Spoken%20Dialogue%20Systems.%20Preprint%20at%20%3Ca%20class%3D%27zp-DOIURL%27%20href%3D%27https%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2503.08533%27%3Ehttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2503.08533%3C%5C%2Fa%3E%20%282025%29.%3C%5C%2Fdiv%3E%5Cn%20%20%3C%5C%2Fdiv%3E%5Cn%3C%5C%2Fdiv%3E%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22ESPnet-SDS%3A%20Unified%20Toolkit%20and%20Demo%20for%20Spoken%20Dialogue%20Systems%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Siddhant%22%2C%22lastName%22%3A%22Arora%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Yifan%22%2C%22lastName%22%3A%22Peng%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Jiatong%22%2C%22lastName%22%3A%22Shi%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Jinchuan%22%2C%22lastName%22%3A%22Tian%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22William%22%2C%22lastName%22%3A%22Chen%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Shikhar%22%2C%22lastName%22%3A%22Bharadwaj%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Hayato%22%2C%22lastName%22%3A%22Futami%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Yosuke%22%2C%22lastName%22%3A%22Kashiwagi%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Emiru%22%2C%22lastName%22%3A%22Tsunoo%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Shuichiro%22%2C%22lastName%22%3A%22Shimizu%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Vaibhav%22%2C%22lastName%22%3A%22Srivastav%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Shinji%22%2C%22lastName%22%3A%22Watanabe%22%7D%5D%2C%22abstractNote%22%3A%22Advancements%20in%20audio%20foundation%20models%20%28FMs%29%20have%20fueled%20interest%20in%20end-to-end%20%28E2E%29%20spoken%20dialogue%20systems%2C%20but%20different%20web%20interfaces%20for%20each%20system%20makes%20it%20challenging%20to%20compare%20and%20contrast%20them%20effectively.%20Motivated%20by%20this%2C%20we%20introduce%20an%20open-source%2C%20user-friendly%20toolkit%20designed%20to%20build%20unified%20web%20interfaces%20for%20various%20cascaded%20and%20E2E%20spoken%20dialogue%20systems.%20Our%20demo%20further%20provides%20users%20with%20the%20option%20to%20get%20on-the-fly%20automated%20evaluation%20metrics%20such%20as%20%281%29%20latency%2C%20%282%29%20ability%20to%20understand%20user%20input%2C%20%283%29%20coherence%2C%20diversity%2C%20and%20relevance%20of%20system%20response%2C%20and%20%284%29%20intelligibility%20and%20audio%20quality%20of%20system%20output.%20Using%20the%20evaluation%20metrics%2C%20we%20compare%20various%20cascaded%20and%20E2E%20spoken%20dialogue%20systems%20with%20a%20human-human%20conversation%20dataset%20as%20a%20proxy.%20Our%20analysis%20demonstrates%20that%20the%20toolkit%20allows%20researchers%20to%20effortlessly%20compare%20and%20contrast%20different%20technologies%2C%20providing%20valuable%20insights%20such%20as%20current%20E2E%20systems%20having%20poorer%20audio%20quality%20and%20less%20diverse%20responses.%20An%20example%20demo%20produced%20using%20our%20toolkit%20is%20publicly%20available%20here%3A%20https%3A%5C%2F%5C%2Fhuggingface.co%5C%2Fspaces%5C%2FSiddhant%5C%2FVoice_Assistant_Demo.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222025%22%2C%22DOI%22%3A%2210.48550%5C%2FARXIV.2503.08533%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2503.08533%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222025-03-17T15%3A35%3A52Z%22%7D%7D%2C%7B%22key%22%3A%2232Z5YN4Q%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Wilfong%20et%20al.%22%2C%22parsedDate%22%3A%222025%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%3Cdiv%20class%3D%5C%22csl-bib-body%5C%22%20style%3D%5C%22line-height%3A%202%3B%20%5C%22%3E%5Cn%20%20%3Cdiv%20class%3D%5C%22csl-entry%5C%22%20style%3D%5C%22clear%3A%20left%3B%20%5C%22%3E%5Cn%20%20%20%20%3Cdiv%20class%3D%5C%22csl-left-margin%5C%22%20style%3D%5C%22float%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%5C%22%3E1.%3C%5C%2Fdiv%3E%3Cdiv%20class%3D%5C%22csl-right-inline%5C%22%20style%3D%5C%22margin%3A%200%20.4em%200%201.5em%3B%5C%22%3EWilfong%2C%20B.%20%3Ci%3Eet%20al.%3C%5C%2Fi%3E%20MFC%205.0%3A%20An%20exascale%20many-physics%20flow%20solver.%20Preprint%20at%20%3Ca%20class%3D%27zp-DOIURL%27%20href%3D%27https%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2503.07953%27%3Ehttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2503.07953%3C%5C%2Fa%3E%20%282025%29.%3C%5C%2Fdiv%3E%5Cn%20%20%3C%5C%2Fdiv%3E%5Cn%3C%5C%2Fdiv%3E%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22MFC%205.0%3A%20An%20exascale%20many-physics%20flow%20solver%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Benjamin%22%2C%22lastName%22%3A%22Wilfong%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Henry%20A.%20Le%22%2C%22lastName%22%3A%22Berre%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Anand%22%2C%22lastName%22%3A%22Radhakrishnan%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Ansh%22%2C%22lastName%22%3A%22Gupta%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Diego%22%2C%22lastName%22%3A%22Vaca-Revelo%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Dimitrios%22%2C%22lastName%22%3A%22Adam%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Haocheng%22%2C%22lastName%22%3A%22Yu%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Hyeoksu%22%2C%22lastName%22%3A%22Lee%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Jose%20Rodolfo%22%2C%22lastName%22%3A%22Chreim%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Mirelys%20Carcana%22%2C%22lastName%22%3A%22Barbosa%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Yanjun%22%2C%22lastName%22%3A%22Zhang%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Esteban%22%2C%22lastName%22%3A%22Cisneros-Garibay%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Aswin%22%2C%22lastName%22%3A%22Gnanaskandan%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Mauro%22%2C%22lastName%22%3A%22Rodriguez%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Reuben%20D.%22%2C%22lastName%22%3A%22Budiardja%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Stephen%22%2C%22lastName%22%3A%22Abbott%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Tim%22%2C%22lastName%22%3A%22Colonius%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Spencer%20H.%22%2C%22lastName%22%3A%22Bryngelson%22%7D%5D%2C%22abstractNote%22%3A%22Engineering%2C%20medicine%2C%20and%20the%20fundamental%20sciences%20broadly%20rely%20on%20flow%20simulations%2C%20making%20performant%20computational%20fluid%20dynamics%20solvers%20an%20open%20source%20software%20mainstay.%20A%20previous%20work%20made%20MFC%203.0%20a%20published%20open%20source%20source%20solver%20with%20many%20features.%20MFC%205.0%20is%20a%20marked%20update%20to%20MFC%203.0%2C%20including%20a%20broad%20set%20of%20well-established%20and%20novel%20physical%20models%20and%20numerical%20methods%20and%20the%20introduction%20of%20GPU%20and%20APU%20%28or%20superchip%29%20acceleration.%20We%20exhibit%20state-of-the-art%20performance%20and%20ideal%20scaling%20on%20the%20first%20two%20exascale%20supercomputers%2C%20OLCF%20Frontier%20and%20LLNL%20El%20Capitan.%20Combined%20with%20MFC%27s%20single-GPU%5C%2FAPU%20performance%2C%20MFC%20achieves%20exascale%20computation%20in%20practice.%20With%20these%20capabilities%2C%20MFC%20has%20evolved%20into%20a%20tool%20for%20conducting%20simulations%20that%20many%20engineering%20challenge%20problems%20hinge%20upon.%20New%20physical%20features%20include%20the%20immersed%20boundary%20method%2C%20%24N%24-fluid%20phase%20change%2C%20Euler--Euler%20and%20Euler--Lagrange%20sub-grid%20bubble%20models%2C%20fluid-structure%20interaction%2C%20hypo-%20and%20hyper-elastic%20materials%2C%20chemically%20reacting%20flow%2C%20two-material%20surface%20tension%2C%20and%20more.%20Numerical%20techniques%20now%20represent%20the%20current%20state-of-the-art%2C%20including%20general%20relaxation%20characteristic%20boundary%20conditions%2C%20WENO%20variants%2C%20Strang%20splitting%20for%20stiff%20sub-grid%20flow%20features%2C%20and%20low%20Mach%20number%20treatments.%20Weak%20scaling%20to%20tens%20of%20thousands%20of%20GPUs%20on%20OLCF%20Frontier%20and%20LLNL%20El%20Capitan%20see%20efficiencies%20within%205%25%20of%20ideal%20to%20over%2090%25%20of%20their%20respective%20system%20sizes.%20Strong%20scaling%20results%20for%20a%2016-time%20increase%20in%20device%20count%20show%20parallel%20efficiencies%20over%2090%25%20on%20OLCF%20Frontier.%20Other%20MFC%20improvements%20include%20ensuring%20code%20resilience%20and%20correctness%20with%20a%20continuous%20integration%20suite%2C%20the%20use%20of%20metaprogramming%20to%20reduce%20code%20length%20and%20maintain%20performance%20portability%2C%20and%20efficient%20computational%20representations%20for%20chemical%20reactions%20and%20thermodynamics%20via%20code%20generation%20with%20Pyrometheus.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222025%22%2C%22DOI%22%3A%2210.48550%5C%2FARXIV.2503.07953%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2503.07953%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222025-03-17T15%3A30%3A18Z%22%7D%7D%2C%7B%22key%22%3A%22CJUS7K2Z%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Kearns%20et%20al.%22%2C%22parsedDate%22%3A%222025-03-11%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%3Cdiv%20class%3D%5C%22csl-bib-body%5C%22%20style%3D%5C%22line-height%3A%202%3B%20%5C%22%3E%5Cn%20%20%3Cdiv%20class%3D%5C%22csl-entry%5C%22%20style%3D%5C%22clear%3A%20left%3B%20%5C%22%3E%5Cn%20%20%20%20%3Cdiv%20class%3D%5C%22csl-left-margin%5C%22%20style%3D%5C%22float%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%5C%22%3E1.%3C%5C%2Fdiv%3E%3Cdiv%20class%3D%5C%22csl-right-inline%5C%22%20style%3D%5C%22margin%3A%200%20.4em%200%201.5em%3B%5C%22%3EKearns%2C%20F.%20L.%20%3Ci%3Eet%20al.%3C%5C%2Fi%3E%20D614G%20reshapes%20allosteric%20networks%20and%20opening%20mechanisms%20of%20SARS-CoV-2%20spikes.%20Preprint%20at%20%3Ca%20class%3D%27zp-DOIURL%27%20href%3D%27https%3A%5C%2F%5C%2Fdoi.org%5C%2F10.1101%5C%2F2025.03.07.642081%27%3Ehttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.1101%5C%2F2025.03.07.642081%3C%5C%2Fa%3E%20%282025%29.%3C%5C%2Fdiv%3E%5Cn%20%20%3C%5C%2Fdiv%3E%5Cn%3C%5C%2Fdiv%3E%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22D614G%20reshapes%20allosteric%20networks%20and%20opening%20mechanisms%20of%20SARS-CoV-2%20spikes%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Fiona%20L.%22%2C%22lastName%22%3A%22Kearns%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Anthony%20T.%22%2C%22lastName%22%3A%22Bogetti%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Carla%22%2C%22lastName%22%3A%22Calv%5Cu00f3-Tusell%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Mac%20Kevin%20E.%22%2C%22lastName%22%3A%22Braza%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Lorenzo%22%2C%22lastName%22%3A%22Casalino%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Amanda%20J.%22%2C%22lastName%22%3A%22Gramm%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Sean%22%2C%22lastName%22%3A%22Braet%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Mia%20A.%22%2C%22lastName%22%3A%22Rosenfeld%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Harinda%22%2C%22lastName%22%3A%22Rajapaksha%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Bryan%22%2C%22lastName%22%3A%22Barker%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Ganesh%22%2C%22lastName%22%3A%22Anand%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Surl-Hee%22%2C%22lastName%22%3A%22Ahn%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Lillian%20T.%22%2C%22lastName%22%3A%22Chong%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Rommie%20E.%22%2C%22lastName%22%3A%22Amaro%22%7D%5D%2C%22abstractNote%22%3A%22Abstract%20%5Cn%20%20%20%20%20%20%20%20%20%20The%20SARS-CoV-2%20spike%20glycoprotein%20binds%20human%20epithelial%20cells%20and%20enables%20infection%20through%20a%20key%20conformational%20transition%20that%20exposes%20its%20receptor%20binding%20domain%20%28RBD%29.%20Experimental%20evidence%20indicates%20that%20spike%20mutations%2C%20particularly%20the%20early%20D614G%20variant%2C%20alter%20the%20rate%20of%20this%20conformational%20shift%2C%20potentially%20increasing%20viral%20infectivity.%20To%20investigate%20how%20mutations%20reshape%20the%20conformational%20landscape%2C%20we%20conducted%20extensive%20weighted%20ensemble%20simulations%20of%20the%20Ancestral%2C%20Delta%2C%20and%20Omicron%20BA.1%20spike%20strains%20along%20the%20RBD%20opening%20pathway.%20We%20observe%20that%20Ancestral%2C%20Delta%2C%20and%20Omicron%20BA.1%20spike%20RBDs%20open%20differently%2C%20with%20Omicron%20BA.1%20following%20a%20more%20direct%20opening%20profile%20until%20it%20reaches%20a%20%5Cu201csuper-open%5Cu201d%20state%20wherein%20it%20begins%20to%20%5Cu201cpeel%5Cu201d%2C%20suggesting%20increased%20S1%20flexibility.%20Via%20dynamical%20network%20analysis%2C%20we%20identified%20two%20allosteric%20communication%20networks%20uniting%20all%20S1%20domains%3A%20the%20established%20N2R%20linker%20and%20a%20newly%20discovered%20anti-parallel%20R2N%20linker.%20In%20Delta%20and%20Omicron%20BA.1%20variant%20spikes%2C%20RBD%20opening%20is%20facilitated%20by%20both%20linkers%2C%20while%20the%20Ancestral%20strain%20relies%20predominantly%20on%20the%20N2R%20linker.%20In%20the%20ancestral%20spike%2C%20the%20D614-K854%20salt%20bridge%20impedes%20allosteric%20communication%20through%20the%20R2N%20linker%2C%20whereas%20the%20loss%20of%20this%20salt%20bridge%20in%20all%20subsequent%20VOCs%20alleviates%20local%20frustration%20and%2C%20we%20believe%2C%20accelerates%20RBD%20opening.%20Hydrogen-deuterium%20mass%20spectrometry%20experiments%20validate%20these%20altered%20dynamics%20in%20the%20D614%20region%20across%20Ancestral%2C%20D614G%2C%20and%20Omicron%20BA.1%20spikes.%20This%20study%20unveils%20a%20%5Cu2018hidden%5Cu2019%20allosteric%20network%2C%20connecting%20the%20NTD%20to%20the%20RBD%20via%20the%20614-proximal%20region%2C%20and%20the%20D614G%20mutation%20reshapes%20the%20fitness%20landscape%20of%20these%20critical%20viral%20glycoproteins.%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20Significance%20Statement%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20Our%20work%20reveals%20how%20the%20D614G%20mutation%20in%20the%20SARS-CoV-2%20spike%20protein%20reshapes%20its%20internal%20communication%20pathways%20and%20speeds%20up%20receptor%20binding%20domain%20%28RBD%29%20opening%2C%20providing%20mechanistic%20insight%20into%20the%20evolution%20and%20enhanced%20infectivity%20of%20SARS-CoV-2%20variants%20of%20concern.%20We%20also%20describe%20differences%20in%20opening%20pathways%20and%20relative%20rates%20of%20opening%20for%20Delta%20and%20Omicron%20BA.1%20spike%20RBDs%20relative%20to%20the%20original%20%28Ancestral%29%20coronavirus%20strain%20from%20Wuhan%2C%20China.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222025-03-11%22%2C%22DOI%22%3A%2210.1101%5C%2F2025.03.07.642081%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22http%3A%5C%2F%5C%2Fbiorxiv.org%5C%2Flookup%5C%2Fdoi%5C%2F10.1101%5C%2F2025.03.07.642081%22%2C%22language%22%3A%22en%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222025-03-13T19%3A06%3A32Z%22%7D%7D%2C%7B%22key%22%3A%22U7EZQFS5%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Yang%20et%20al.%22%2C%22parsedDate%22%3A%222025-04-06%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%3Cdiv%20class%3D%5C%22csl-bib-body%5C%22%20style%3D%5C%22line-height%3A%202%3B%20%5C%22%3E%5Cn%20%20%3Cdiv%20class%3D%5C%22csl-entry%5C%22%20style%3D%5C%22clear%3A%20left%3B%20%5C%22%3E%5Cn%20%20%20%20%3Cdiv%20class%3D%5C%22csl-left-margin%5C%22%20style%3D%5C%22float%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%5C%22%3E1.%3C%5C%2Fdiv%3E%3Cdiv%20class%3D%5C%22csl-right-inline%5C%22%20style%3D%5C%22margin%3A%200%20.4em%200%201.5em%3B%5C%22%3EYang%2C%20Y.%2C%20Taherian%2C%20H.%2C%20Kalkhoriani%2C%20V.%20A.%20%26amp%3B%20Wang%2C%20D.%20Elevating%20Robust%20ASR%20By%20Decoupling%20Multi-Channel%20Speaker%20Separation%20and%20Speech%20Recognition.%20in%20%3Ci%3EICASSP%202025%20-%202025%20IEEE%20International%20Conference%20on%20Acoustics%2C%20Speech%20and%20Signal%20Processing%20%28ICASSP%29%3C%5C%2Fi%3E%20%28IEEE%2C%20Hyderabad%2C%20India%2C%202025%29.%20doi%3Ahttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.1109%5C%2FICASSP49660.2025.10888074.%3C%5C%2Fdiv%3E%5Cn%20%20%3C%5C%2Fdiv%3E%5Cn%3C%5C%2Fdiv%3E%22%2C%22data%22%3A%7B%22itemType%22%3A%22conferencePaper%22%2C%22title%22%3A%22Elevating%20Robust%20ASR%20By%20Decoupling%20Multi-Channel%20Speaker%20Separation%20and%20Speech%20Recognition%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Yufeng%22%2C%22lastName%22%3A%22Yang%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Hassan%22%2C%22lastName%22%3A%22Taherian%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Vahid%20Ahmadi%22%2C%22lastName%22%3A%22Kalkhoriani%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22DiLiang%22%2C%22lastName%22%3A%22Wang%22%7D%5D%2C%22abstractNote%22%3A%22%22%2C%22date%22%3A%222025-04-06%22%2C%22proceedingsTitle%22%3A%22ICASSP%202025%20-%202025%20IEEE%20International%20Conference%20on%20Acoustics%2C%20Speech%20and%20Signal%20Processing%20%28ICASSP%29%22%2C%22conferenceName%22%3A%22ICASSP%202025%20-%202025%20IEEE%20International%20Conference%20on%20Acoustics%2C%20Speech%20and%20Signal%20Processing%20%28ICASSP%29%22%2C%22language%22%3A%22%22%2C%22DOI%22%3A%22https%3A%5C%2F%5C%2Fdoi.org%5C%2F10.1109%5C%2FICASSP49660.2025.10888074%22%2C%22ISBN%22%3A%22979-8-3503-6874-1%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Fieeexplore.ieee.org%5C%2Fabstract%5C%2Fdocument%5C%2F10888074%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222025-03-11T13%3A55%3A51Z%22%7D%7D%2C%7B%22key%22%3A%22TF26SQG4%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Chung%20et%20al.%22%2C%22parsedDate%22%3A%222024-12-15%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%3Cdiv%20class%3D%5C%22csl-bib-body%5C%22%20style%3D%5C%22line-height%3A%202%3B%20%5C%22%3E%5Cn%20%20%3Cdiv%20class%3D%5C%22csl-entry%5C%22%20style%3D%5C%22clear%3A%20left%3B%20%5C%22%3E%5Cn%20%20%20%20%3Cdiv%20class%3D%5C%22csl-left-margin%5C%22%20style%3D%5C%22float%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%5C%22%3E1.%3C%5C%2Fdiv%3E%3Cdiv%20class%3D%5C%22csl-right-inline%5C%22%20style%3D%5C%22margin%3A%200%20.4em%200%201.5em%3B%5C%22%3EChung%2C%20J.%2C%20Zhang%2C%20C.%20%26amp%3B%20Chen%2C%20T.%20Mobility%20Scooter%20Riding%20Behavior%20Stability%20Analysis%20Based%20on%20Multimodal%20Contrastive%20Learning.%20in%20%3Ci%3E2024%20IEEE%20International%20Conference%20on%20Big%20Data%20%28BigData%29%3C%5C%2Fi%3E%206439%26%23x2013%3B6445%20%28IEEE%2C%20Washington%2C%20DC%2C%20USA%2C%202024%29.%20%3Ca%20class%3D%27zp-DOIURL%27%20href%3D%27http%3A%5C%2F%5C%2Fdoi.org%5C%2F10.1109%5C%2FBigData62323.2024.10825478%27%3Ehttp%3A%5C%2F%5C%2Fdoi.org%5C%2F10.1109%5C%2FBigData62323.2024.10825478%3C%5C%2Fa%3E.%3C%5C%2Fdiv%3E%5Cn%20%20%3C%5C%2Fdiv%3E%5Cn%3C%5C%2Fdiv%3E%22%2C%22data%22%3A%7B%22itemType%22%3A%22conferencePaper%22%2C%22title%22%3A%22Mobility%20Scooter%20Riding%20Behavior%20Stability%20Analysis%20Based%20on%20Multimodal%20Contrastive%20Learning%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Justin%22%2C%22lastName%22%3A%22Chung%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Chenrui%22%2C%22lastName%22%3A%22Zhang%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Tingting%22%2C%22lastName%22%3A%22Chen%22%7D%5D%2C%22abstractNote%22%3A%22%22%2C%22date%22%3A%222024-12-15%22%2C%22proceedingsTitle%22%3A%222024%20IEEE%20International%20Conference%20on%20Big%20Data%20%28BigData%29%22%2C%22conferenceName%22%3A%222024%20IEEE%20International%20Conference%20on%20Big%20Data%20%28BigData%29%22%2C%22language%22%3A%22%22%2C%22DOI%22%3A%2210.1109%5C%2FBigData62323.2024.10825478%22%2C%22ISBN%22%3A%229798350362480%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Fieeexplore.ieee.org%5C%2Fdocument%5C%2F10825478%5C%2F%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222025-03-10T17%3A19%3A58Z%22%7D%7D%2C%7B%22key%22%3A%225NEQ5TK9%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Hossain%20et%20al.%22%2C%22parsedDate%22%3A%222025-03-06%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%3Cdiv%20class%3D%5C%22csl-bib-body%5C%22%20style%3D%5C%22line-height%3A%202%3B%20%5C%22%3E%5Cn%20%20%3Cdiv%20class%3D%5C%22csl-entry%5C%22%20style%3D%5C%22clear%3A%20left%3B%20%5C%22%3E%5Cn%20%20%20%20%3Cdiv%20class%3D%5C%22csl-left-margin%5C%22%20style%3D%5C%22float%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%5C%22%3E1.%3C%5C%2Fdiv%3E%3Cdiv%20class%3D%5C%22csl-right-inline%5C%22%20style%3D%5C%22margin%3A%200%20.4em%200%201.5em%3B%5C%22%3EHossain%2C%20R.%20%3Ci%3Eet%20al.%3C%5C%2Fi%3E%20Virtual%20sensing-enabled%20digital%20twin%20framework%20for%20real-time%20monitoring%20of%20nuclear%20systems%20leveraging%20deep%20neural%20operators.%20%3Ci%3Enpj%20Mater%20Degrad%3C%5C%2Fi%3E%20%3Cb%3E9%3C%5C%2Fb%3E%2C%2021%20%282025%29.%3C%5C%2Fdiv%3E%5Cn%20%20%3C%5C%2Fdiv%3E%5Cn%3C%5C%2Fdiv%3E%22%2C%22data%22%3A%7B%22itemType%22%3A%22journalArticle%22%2C%22title%22%3A%22Virtual%20sensing-enabled%20digital%20twin%20framework%20for%20real-time%20monitoring%20of%20nuclear%20systems%20leveraging%20deep%20neural%20operators%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Raisa%22%2C%22lastName%22%3A%22Hossain%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Farid%22%2C%22lastName%22%3A%22Ahmed%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Kazuma%22%2C%22lastName%22%3A%22Kobayashi%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Seid%22%2C%22lastName%22%3A%22Koric%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Diab%22%2C%22lastName%22%3A%22Abueidda%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Syed%20Bahauddin%22%2C%22lastName%22%3A%22Alam%22%7D%5D%2C%22abstractNote%22%3A%22Abstract%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20Real-time%20monitoring%20is%20a%20foundation%20of%20nuclear%20digital%20twin%20technology%2C%20crucial%20for%20detecting%20material%20degradation%20and%20maintaining%20nuclear%20system%20integrity.%20Traditional%20physical%20sensor%20systems%20face%20limitations%2C%20particularly%20in%20measuring%20critical%20parameters%20in%20hard-to-reach%20or%20harsh%20environments%2C%20often%20resulting%20in%20incomplete%20data%20coverage.%20Machine%20learning-driven%20virtual%20sensors%20offer%20a%20transformative%20solution%20by%20complementing%20physical%20sensors%20in%20monitoring%20critical%20degradation%20indicators.%20This%20paper%20introduces%20the%20use%20of%20Deep%20Operator%20Networks%20%28DeepONet%29%20to%20predict%20key%20thermal-hydraulic%20parameters%20in%20the%20hot%20leg%20of%20pressurized%20water%20reactor.%20DeepONet%20acts%20as%20a%20virtual%20sensor%2C%20mapping%20operational%20inputs%20to%20spatially%20distributed%20system%20behaviors%20without%20requiring%20frequent%20retraining.%20Our%20results%20show%20that%20DeepONet%20achieves%20low%20mean%20squared%20and%20Relative%20L2%20error%2C%20making%20predictions%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%201400%20times%20faster%20than%20traditional%20CFD%20simulations%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20.%20These%20characteristics%20enable%20DeepONet%20to%20function%20as%20a%20real-time%20virtual%20sensor%2C%20synchronizing%20with%20the%20physical%20system%20to%20track%20degradation%20conditions%20and%20provide%20insights%20within%20the%20digital%20twin%20framework%20for%20nuclear%20systems.%22%2C%22date%22%3A%222025-03-06%22%2C%22language%22%3A%22en%22%2C%22DOI%22%3A%2210.1038%5C%2Fs41529-025-00557-y%22%2C%22ISSN%22%3A%222397-2106%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Fwww.nature.com%5C%2Farticles%5C%2Fs41529-025-00557-y%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222025-03-10T16%3A41%3A00Z%22%7D%7D%2C%7B%22key%22%3A%22N8PMXP3Y%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Pulavarthi%20et%20al.%22%2C%22parsedDate%22%3A%222025%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%3Cdiv%20class%3D%5C%22csl-bib-body%5C%22%20style%3D%5C%22line-height%3A%202%3B%20%5C%22%3E%5Cn%20%20%3Cdiv%20class%3D%5C%22csl-entry%5C%22%20style%3D%5C%22clear%3A%20left%3B%20%5C%22%3E%5Cn%20%20%20%20%3Cdiv%20class%3D%5C%22csl-left-margin%5C%22%20style%3D%5C%22float%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%5C%22%3E1.%3C%5C%2Fdiv%3E%3Cdiv%20class%3D%5C%22csl-right-inline%5C%22%20style%3D%5C%22margin%3A%200%20.4em%200%201.5em%3B%5C%22%3EPulavarthi%2C%20V.%2C%20Nandal%2C%20D.%2C%20Dan%2C%20S.%20%26amp%3B%20Pal%2C%20D.%20Are%20LLMs%20Ready%20for%20Practical%20Adoption%20for%20Assertion%20Generation%3F%20Preprint%20at%20%3Ca%20class%3D%27zp-DOIURL%27%20href%3D%27https%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2502.20633%27%3Ehttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2502.20633%3C%5C%2Fa%3E%20%282025%29.%3C%5C%2Fdiv%3E%5Cn%20%20%3C%5C%2Fdiv%3E%5Cn%3C%5C%2Fdiv%3E%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22Are%20LLMs%20Ready%20for%20Practical%20Adoption%20for%20Assertion%20Generation%3F%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Vaishnavi%22%2C%22lastName%22%3A%22Pulavarthi%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Deeksha%22%2C%22lastName%22%3A%22Nandal%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Soham%22%2C%22lastName%22%3A%22Dan%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Debjit%22%2C%22lastName%22%3A%22Pal%22%7D%5D%2C%22abstractNote%22%3A%22Assertions%20have%20been%20the%20de%20facto%20collateral%20for%20simulation-based%20and%20formal%20verification%20of%20hardware%20designs%20for%20over%20a%20decade.%20The%20quality%20of%20hardware%20verification%2C%20i.e.%2C%20detection%20and%20diagnosis%20of%20corner-case%20design%20bugs%2C%20is%20critically%20dependent%20on%20the%20quality%20of%20the%20assertions.%20With%20the%20onset%20of%20generative%20AI%20such%20as%20Transformers%20and%20Large-Language%20Models%20%28LLMs%29%2C%20there%20has%20been%20a%20renewed%20interest%20in%20developing%20novel%2C%20effective%2C%20and%20scalable%20techniques%20of%20generating%20functional%20and%20security%20assertions%20from%20design%20source%20code.%20While%20there%20have%20been%20recent%20works%20that%20use%20commercial-of-the-shelf%20%28COTS%29%20LLMs%20for%20assertion%20generation%2C%20there%20is%20no%20comprehensive%20study%20in%20quantifying%20the%20effectiveness%20of%20LLMs%20in%20generating%20syntactically%20and%20semantically%20correct%20assertions.%20In%20this%20paper%2C%20we%20first%20discuss%20AssertionBench%20from%20our%20prior%20work%2C%20a%20comprehensive%20set%20of%20designs%20and%20assertions%20to%20quantify%20the%20goodness%20of%20a%20broad%20spectrum%20of%20COTS%20LLMs%20for%20the%20task%20of%20assertion%20generations%20from%20hardware%20design%20source%20code.%20Our%20key%20insight%20was%20that%20COTS%20LLMs%20are%20not%20yet%20ready%20for%20prime-time%20adoption%20for%20assertion%20generation%20as%20they%20generate%20a%20considerable%20fraction%20of%20syntactically%20and%20semantically%20incorrect%20assertions.%20Motivated%20by%20the%20insight%2C%20we%20propose%20AssertionLLM%2C%20a%20first%20of%20its%20kind%20LLM%20model%2C%20specifically%20fine-tuned%20for%20assertion%20generation.%20Our%20initial%20experimental%20results%20show%20that%20AssertionLLM%20considerably%20improves%20the%20semantic%20and%20syntactic%20correctness%20of%20the%20generated%20assertions%20over%20COTS%20LLMs.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222025%22%2C%22DOI%22%3A%2210.48550%5C%2FARXIV.2502.20633%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2502.20633%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222025-03-10T16%3A30%3A31Z%22%7D%7D%2C%7B%22key%22%3A%22229MBA5P%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Marques%20et%20al.%22%2C%22parsedDate%22%3A%222025%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%3Cdiv%20class%3D%5C%22csl-bib-body%5C%22%20style%3D%5C%22line-height%3A%202%3B%20%5C%22%3E%5Cn%20%20%3Cdiv%20class%3D%5C%22csl-entry%5C%22%20style%3D%5C%22clear%3A%20left%3B%20%5C%22%3E%5Cn%20%20%20%20%3Cdiv%20class%3D%5C%22csl-left-margin%5C%22%20style%3D%5C%22float%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%5C%22%3E1.%3C%5C%2Fdiv%3E%3Cdiv%20class%3D%5C%22csl-right-inline%5C%22%20style%3D%5C%22margin%3A%200%20.4em%200%201.5em%3B%5C%22%3EMarques%2C%20J.%20M.%20C.%20%3Ci%3Eet%20al.%3C%5C%2Fi%3E%20Map%20Space%20Belief%20Prediction%20for%20Manipulation-Enhanced%20Mapping.%20Preprint%20at%20%3Ca%20class%3D%27zp-DOIURL%27%20href%3D%27https%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2502.20606%27%3Ehttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2502.20606%3C%5C%2Fa%3E%20%282025%29.%3C%5C%2Fdiv%3E%5Cn%20%20%3C%5C%2Fdiv%3E%5Cn%3C%5C%2Fdiv%3E%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22Map%20Space%20Belief%20Prediction%20for%20Manipulation-Enhanced%20Mapping%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Joao%20Marcos%20Correia%22%2C%22lastName%22%3A%22Marques%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Nils%22%2C%22lastName%22%3A%22Dengler%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Tobias%22%2C%22lastName%22%3A%22Zaenker%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Jesper%22%2C%22lastName%22%3A%22Mucke%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Shenlong%22%2C%22lastName%22%3A%22Wang%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Maren%22%2C%22lastName%22%3A%22Bennewitz%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Kris%22%2C%22lastName%22%3A%22Hauser%22%7D%5D%2C%22abstractNote%22%3A%22Searching%20for%20objects%20in%20cluttered%20environments%20requires%20selecting%20efficient%20viewpoints%20and%20manipulation%20actions%20to%20remove%20occlusions%20and%20reduce%20uncertainty%20in%20object%20locations%2C%20shapes%2C%20and%20categories.%20In%20this%20work%2C%20we%20address%20the%20problem%20of%20manipulation-enhanced%20semantic%20mapping%2C%20where%20a%20robot%20has%20to%20efficiently%20identify%20all%20objects%20in%20a%20cluttered%20shelf.%20Although%20Partially%20Observable%20Markov%20Decision%20Processes~%28POMDPs%29%20are%20standard%20for%20decision-making%20under%20uncertainty%2C%20representing%20unstructured%20interactive%20worlds%20remains%20challenging%20in%20this%20formalism.%20To%20tackle%20this%2C%20we%20define%20a%20POMDP%20whose%20belief%20is%20summarized%20by%20a%20metric-semantic%20grid%20map%20and%20propose%20a%20novel%20framework%20that%20uses%20neural%20networks%20to%20perform%20map-space%20belief%20updates%20to%20reason%20efficiently%20and%20simultaneously%20about%20object%20geometries%2C%20locations%2C%20categories%2C%20occlusions%2C%20and%20manipulation%20physics.%20Further%2C%20to%20enable%20accurate%20information%20gain%20analysis%2C%20the%20learned%20belief%20updates%20should%20maintain%20calibrated%20estimates%20of%20uncertainty.%20Therefore%2C%20we%20propose%20Calibrated%20Neural-Accelerated%20Belief%20Updates%20%28CNABUs%29%20to%20learn%20a%20belief%20propagation%20model%20that%20generalizes%20to%20novel%20scenarios%20and%20provides%20confidence-calibrated%20predictions%20for%20unknown%20areas.%20Our%20experiments%20show%20that%20our%20novel%20POMDP%20planner%20improves%20map%20completeness%20and%20accuracy%20over%20existing%20methods%20in%20challenging%20simulations%20and%20successfully%20transfers%20to%20real-world%20cluttered%20shelves%20in%20zero-shot%20fashion.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222025%22%2C%22DOI%22%3A%2210.48550%5C%2FARXIV.2502.20606%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2502.20606%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222025-03-10T16%3A22%3A10Z%22%7D%7D%2C%7B%22key%22%3A%22IKLS3D6J%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Singer%20et%20al.%22%2C%22parsedDate%22%3A%222025%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%3Cdiv%20class%3D%5C%22csl-bib-body%5C%22%20style%3D%5C%22line-height%3A%202%3B%20%5C%22%3E%5Cn%20%20%3Cdiv%20class%3D%5C%22csl-entry%5C%22%20style%3D%5C%22clear%3A%20left%3B%20%5C%22%3E%5Cn%20%20%20%20%3Cdiv%20class%3D%5C%22csl-left-margin%5C%22%20style%3D%5C%22float%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%5C%22%3E1.%3C%5C%2Fdiv%3E%3Cdiv%20class%3D%5C%22csl-right-inline%5C%22%20style%3D%5C%22margin%3A%200%20.4em%200%201.5em%3B%5C%22%3ESinger%2C%20L.%20P.%20%3Ci%3Eet%20al.%3C%5C%2Fi%3E%20Optimal%20Follow-Up%20of%20Gravitational-Wave%20Events%20with%20the%20UltraViolet%20EXplorer%20%28UVEX%29.%20Preprint%20at%20%3Ca%20class%3D%27zp-DOIURL%27%20href%3D%27https%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2502.17560%27%3Ehttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2502.17560%3C%5C%2Fa%3E%20%282025%29.%3C%5C%2Fdiv%3E%5Cn%20%20%3C%5C%2Fdiv%3E%5Cn%3C%5C%2Fdiv%3E%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22Optimal%20Follow-Up%20of%20Gravitational-Wave%20Events%20with%20the%20UltraViolet%20EXplorer%20%28UVEX%29%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Leo%20P.%22%2C%22lastName%22%3A%22Singer%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Alexander%20W.%22%2C%22lastName%22%3A%22Criswell%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Sydney%20C.%22%2C%22lastName%22%3A%22Leggio%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22R.%20Weizmann%22%2C%22lastName%22%3A%22Kiendrebeogo%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Michael%20W.%22%2C%22lastName%22%3A%22Coughlin%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Hannah%20P.%22%2C%22lastName%22%3A%22Earnshaw%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Suvi%22%2C%22lastName%22%3A%22Gezari%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Brian%20W.%22%2C%22lastName%22%3A%22Grefenstette%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Fiona%20A.%22%2C%22lastName%22%3A%22Harrison%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Mansi%20M.%22%2C%22lastName%22%3A%22Kasliwal%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Brett%20M.%22%2C%22lastName%22%3A%22Morris%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Erik%22%2C%22lastName%22%3A%22Tollerud%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22S.%20Bradley%22%2C%22lastName%22%3A%22Cenko%22%7D%5D%2C%22abstractNote%22%3A%22The%20UltraViolet%20EXplorer%20%28UVEX%29%20is%20a%20wide-field%20ultraviolet%20space%20telescope%20selected%20as%20a%20NASA%20Medium-Class%20Explorer%20%28MIDEX%29%20mission%20for%20launch%20in%202030.%20UVEX%20will%20undertake%20deep%2C%20cadenced%20surveys%20of%20the%20entire%20sky%20to%20probe%20low%20mass%20galaxies%20and%20explore%20the%20ultraviolet%20%28UV%29%20time-domain%20sky%2C%20and%20it%20will%20carry%20the%20first%20rapidly%20deployable%20UV%20spectroscopic%20capability%20for%20a%20broad%20range%20of%20science%20applications.%20One%20of%20UVEX%27s%20prime%20objectives%20is%20to%20follow%20up%20gravitational%20wave%20%28GW%29%20binary%20neutron%20star%20mergers%20as%20targets%20of%20opportunity%20%28ToOs%29%2C%20rapidly%20scanning%20across%20their%20localization%20regions%20to%20search%20for%20their%20kilonova%20%28KN%29%20counterparts.%20Early-time%20multiband%20ultraviolet%20light%20curves%20of%20KNe%20are%20key%20to%20explaining%20the%20interplay%20between%20jet%20and%20ejecta%20in%20binary%20neutron%20star%20mergers.%20Owing%20to%20high%20Galactic%20extinction%20in%20the%20ultraviolet%20and%20the%20variation%20of%20GW%20distance%20estimates%20over%20the%20sky%2C%20the%20sensitivity%20to%20kilonovae%20can%20vary%20significantly%20across%20the%20GW%20localization%20and%20even%20across%20the%20footprint%20of%20a%20single%20image%20given%20UVEX%27s%20large%20field%20of%20view.%20Good%20ToO%20observing%20strategies%20to%20trade%20off%20between%20area%20and%20depth%20are%20neither%20simple%20nor%20obvious.%20We%20present%20an%20optimal%20strategy%20for%20GW%20follow-up%20with%20UVEX%20in%20which%20exposure%20time%20is%20adjusted%20dynamically%20for%20each%20field%20individually%20to%20maximize%20the%20overall%20probability%20of%20detection.%20We%20model%20the%20scheduling%20problem%20using%20the%20expressive%20and%20powerful%20mathematical%20framework%20of%20mixed%20integer%20linear%20programming%20%28MILP%29%2C%20and%20employ%20a%20state-of-the-art%20MILP%20solver%20to%20automatically%20generate%20observing%20plan%20timelines%20that%20achieve%20high%20probabilities%20of%20kilonova%20detection.%20We%20have%20implemented%20this%20strategy%20in%20an%20open-source%20astronomical%20scheduling%20software%20package%20called%20the%20Multi-Mission%20Multi-Messenger%20Observation%20Planning%20Toolkit%20%28M4OPT%29%2C%20on%20GitHub%20at%20https%3A%5C%2F%5C%2Fgithub.com%5C%2Fm4opt%5C%2Fm4opt.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222025%22%2C%22DOI%22%3A%2210.48550%5C%2FARXIV.2502.17560%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2502.17560%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222025-03-10T16%3A14%3A29Z%22%7D%7D%2C%7B%22key%22%3A%22KV5F87LS%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Tajwar%20et%20al.%22%2C%22parsedDate%22%3A%222025%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%3Cdiv%20class%3D%5C%22csl-bib-body%5C%22%20style%3D%5C%22line-height%3A%202%3B%20%5C%22%3E%5Cn%20%20%3Cdiv%20class%3D%5C%22csl-entry%5C%22%20style%3D%5C%22clear%3A%20left%3B%20%5C%22%3E%5Cn%20%20%20%20%3Cdiv%20class%3D%5C%22csl-left-margin%5C%22%20style%3D%5C%22float%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%5C%22%3E1.%3C%5C%2Fdiv%3E%3Cdiv%20class%3D%5C%22csl-right-inline%5C%22%20style%3D%5C%22margin%3A%200%20.4em%200%201.5em%3B%5C%22%3ETajwar%2C%20F.%20%3Ci%3Eet%20al.%3C%5C%2Fi%3E%20Training%20a%20Generally%20Curious%20Agent.%20Preprint%20at%20%3Ca%20class%3D%27zp-DOIURL%27%20href%3D%27https%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2502.17543%27%3Ehttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2502.17543%3C%5C%2Fa%3E%20%282025%29.%3C%5C%2Fdiv%3E%5Cn%20%20%3C%5C%2Fdiv%3E%5Cn%3C%5C%2Fdiv%3E%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22Training%20a%20Generally%20Curious%20Agent%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Fahim%22%2C%22lastName%22%3A%22Tajwar%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Yiding%22%2C%22lastName%22%3A%22Jiang%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Abitha%22%2C%22lastName%22%3A%22Thankaraj%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Sumaita%20Sadia%22%2C%22lastName%22%3A%22Rahman%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22J%20Zico%22%2C%22lastName%22%3A%22Kolter%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Jeff%22%2C%22lastName%22%3A%22Schneider%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Ruslan%22%2C%22lastName%22%3A%22Salakhutdinov%22%7D%5D%2C%22abstractNote%22%3A%22Efficient%20exploration%20is%20essential%20for%20intelligent%20systems%20interacting%20with%20their%20environment%2C%20but%20existing%20language%20models%20often%20fall%20short%20in%20scenarios%20that%20require%20strategic%20information%20gathering.%20In%20this%20paper%2C%20we%20present%20PAPRIKA%2C%20a%20fine-tuning%20approach%20that%20enables%20language%20models%20to%20develop%20general%20decision-making%20capabilities%20that%20are%20not%20confined%20to%20particular%20environments.%20By%20training%20on%20synthetic%20interaction%20data%20from%20different%20tasks%20that%20require%20diverse%20strategies%2C%20PAPRIKA%20teaches%20models%20to%20explore%20and%20adapt%20their%20behavior%20on%20a%20new%20task%20based%20on%20environment%20feedback%20in-context%20without%20more%20gradient%20updates.%20Experimental%20results%20show%20that%20models%20fine-tuned%20with%20PAPRIKA%20can%20effectively%20transfer%20their%20learned%20decision-making%20capabilities%20to%20entirely%20unseen%20tasks%20without%20additional%20training.%20Unlike%20traditional%20training%2C%20our%20approach%27s%20primary%20bottleneck%20lies%20in%20sampling%20useful%20interaction%20data%20instead%20of%20model%20updates.%20To%20improve%20sample%20efficiency%2C%20we%20propose%20a%20curriculum%20learning%20strategy%20that%20prioritizes%20sampling%20trajectories%20from%20tasks%20with%20high%20learning%20potential.%20These%20results%20suggest%20a%20promising%20path%20towards%20AI%20systems%20that%20can%20autonomously%20solve%20novel%20sequential%20decision-making%20problems%20that%20require%20interactions%20with%20the%20external%20world.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222025%22%2C%22DOI%22%3A%2210.48550%5C%2FARXIV.2502.17543%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2502.17543%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222025-03-10T16%3A04%3A44Z%22%7D%7D%2C%7B%22key%22%3A%22URBFWAUX%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Fiotto-Kaufman%20et%20al.%22%2C%22parsedDate%22%3A%222025-01-03%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%3Cdiv%20class%3D%5C%22csl-bib-body%5C%22%20style%3D%5C%22line-height%3A%202%3B%20%5C%22%3E%5Cn%20%20%3Cdiv%20class%3D%5C%22csl-entry%5C%22%20style%3D%5C%22clear%3A%20left%3B%20%5C%22%3E%5Cn%20%20%20%20%3Cdiv%20class%3D%5C%22csl-left-margin%5C%22%20style%3D%5C%22float%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%5C%22%3E1.%3C%5C%2Fdiv%3E%3Cdiv%20class%3D%5C%22csl-right-inline%5C%22%20style%3D%5C%22margin%3A%200%20.4em%200%201.5em%3B%5C%22%3EFiotto-Kaufman%2C%20J.%20%3Ci%3Eet%20al.%3C%5C%2Fi%3E%20NNsight%20and%20NDIF%3A%20Democratizing%20Access%20to%20Open-Weight%20Foundation%20Model%20Internals.%20Preprint%20at%20%3Ca%20class%3D%27zp-DOIURL%27%20href%3D%27https%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FarXiv.2407.14561%27%3Ehttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FarXiv.2407.14561%3C%5C%2Fa%3E%20%282025%29.%3C%5C%2Fdiv%3E%5Cn%20%20%3C%5C%2Fdiv%3E%5Cn%3C%5C%2Fdiv%3E%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22NNsight%20and%20NDIF%3A%20Democratizing%20Access%20to%20Open-Weight%20Foundation%20Model%20Internals%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Jaden%22%2C%22lastName%22%3A%22Fiotto-Kaufman%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Alexander%20R.%22%2C%22lastName%22%3A%22Loftus%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Eric%22%2C%22lastName%22%3A%22Todd%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Jannik%22%2C%22lastName%22%3A%22Brinkmann%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Koyena%22%2C%22lastName%22%3A%22Pal%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Dmitrii%22%2C%22lastName%22%3A%22Troitskii%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Michael%22%2C%22lastName%22%3A%22Ripa%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Adam%22%2C%22lastName%22%3A%22Belfki%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Can%22%2C%22lastName%22%3A%22Rager%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Caden%22%2C%22lastName%22%3A%22Juang%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Aaron%22%2C%22lastName%22%3A%22Mueller%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Samuel%22%2C%22lastName%22%3A%22Marks%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Arnab%20Sen%22%2C%22lastName%22%3A%22Sharma%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Francesca%22%2C%22lastName%22%3A%22Lucchetti%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Nikhil%22%2C%22lastName%22%3A%22Prakash%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Carla%22%2C%22lastName%22%3A%22Brodley%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Arjun%22%2C%22lastName%22%3A%22Guha%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Jonathan%22%2C%22lastName%22%3A%22Bell%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Byron%20C.%22%2C%22lastName%22%3A%22Wallace%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22David%22%2C%22lastName%22%3A%22Bau%22%7D%5D%2C%22abstractNote%22%3A%22We%20introduce%20NNsight%20and%20NDIF%2C%20technologies%20that%20work%20in%20tandem%20to%20enable%20scientific%20study%20of%20very%20large%20neural%20networks.%20NNsight%20is%20an%20open-source%20system%20that%20extends%20PyTorch%20to%20introduce%20deferred%20remote%20execution.%20NDIF%20is%20a%20scalable%20inference%20service%20that%20executes%20NNsight%20requests%2C%20allowing%20users%20to%20share%20GPU%20resources%20and%20pretrained%20models.%20These%20technologies%20are%20enabled%20by%20the%20intervention%20graph%2C%20an%20architecture%20developed%20to%20decouple%20experiment%20design%20from%20model%20runtime.%20Together%2C%20this%20framework%20provides%20transparent%20and%20efficient%20access%20to%20the%20internals%20of%20deep%20neural%20networks%20such%20as%20very%20large%20language%20models%20%28LLMs%29%20without%20imposing%20the%20cost%20or%20complexity%20of%20hosting%20customized%20models%20individually.%20We%20conduct%20a%20quantitative%20survey%20of%20the%20machine%20learning%20literature%20that%20reveals%20a%20growing%20gap%20in%20the%20study%20of%20the%20internals%20of%20large-scale%20AI.%20We%20demonstrate%20the%20design%20and%20use%20of%20our%20framework%20to%20address%20this%20gap%20by%20enabling%20a%20range%20of%20research%20methods%20on%20huge%20models.%20Finally%2C%20we%20conduct%20benchmarks%20to%20compare%20performance%20with%20previous%20approaches.%20Code%20documentation%2C%20and%20materials%20are%20available%20at%20https%3A%5C%2F%5C%2Fnnsight.net%5C%2F.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22arXiv%3A2407.14561%22%2C%22date%22%3A%222025-01-03%22%2C%22DOI%22%3A%2210.48550%5C%2FarXiv.2407.14561%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22http%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2407.14561%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222025-03-10T16%3A02%3A40Z%22%7D%7D%2C%7B%22key%22%3A%22Q8C6JI65%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Bai%20et%20al.%22%2C%22parsedDate%22%3A%222025-02-13%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%3Cdiv%20class%3D%5C%22csl-bib-body%5C%22%20style%3D%5C%22line-height%3A%202%3B%20%5C%22%3E%5Cn%20%20%3Cdiv%20class%3D%5C%22csl-entry%5C%22%20style%3D%5C%22clear%3A%20left%3B%20%5C%22%3E%5Cn%20%20%20%20%3Cdiv%20class%3D%5C%22csl-left-margin%5C%22%20style%3D%5C%22float%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%5C%22%3E1.%3C%5C%2Fdiv%3E%3Cdiv%20class%3D%5C%22csl-right-inline%5C%22%20style%3D%5C%22margin%3A%200%20.4em%200%201.5em%3B%5C%22%3EBai%2C%20H.%2C%20Zhou%2C%20Y.%2C%20Li%2C%20L.%20E.%2C%20Levine%2C%20S.%20%26amp%3B%20Kumar%2C%20A.%20Digi-Q%3A%20Learning%20Q-Value%20Functions%20for%20Training%20Device-Control%20Agents.%20Preprint%20at%20%3Ca%20class%3D%27zp-DOIURL%27%20href%3D%27https%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FarXiv.2502.15760%27%3Ehttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FarXiv.2502.15760%3C%5C%2Fa%3E%20%282025%29.%3C%5C%2Fdiv%3E%5Cn%20%20%3C%5C%2Fdiv%3E%5Cn%3C%5C%2Fdiv%3E%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22Digi-Q%3A%20Learning%20Q-Value%20Functions%20for%20Training%20Device-Control%20Agents%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Hao%22%2C%22lastName%22%3A%22Bai%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Yifei%22%2C%22lastName%22%3A%22Zhou%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Li%20Erran%22%2C%22lastName%22%3A%22Li%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Sergey%22%2C%22lastName%22%3A%22Levine%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Aviral%22%2C%22lastName%22%3A%22Kumar%22%7D%5D%2C%22abstractNote%22%3A%22While%20a%20number%20of%20existing%20approaches%20for%20building%20foundation%20model%20agents%20rely%20on%20prompting%20or%20fine-tuning%20with%20human%20demonstrations%2C%20it%20is%20not%20sufficient%20in%20dynamic%20environments%20%28e.g.%2C%20mobile%20device%20control%29.%20On-policy%20reinforcement%20learning%20%28RL%29%20should%20address%20these%20limitations%2C%20but%20collecting%20actual%20rollouts%20in%20an%20environment%20is%20often%20undesirable%20in%20truly%20open-ended%20agentic%20problems%20such%20as%20mobile%20device%20control%20or%20interacting%20with%20humans%2C%20where%20each%20unit%20of%20interaction%20is%20associated%20with%20a%20cost.%20In%20such%20scenarios%2C%20a%20method%20for%20policy%20learning%20that%20can%20utilize%20off-policy%20experience%20by%20learning%20a%20trained%20action-value%20function%20is%20much%20more%20effective.%20In%20this%20paper%2C%20we%20develop%20an%20approach%2C%20called%20Digi-Q%2C%20to%20train%20VLM-based%20action-value%20Q-functions%20which%20are%20then%20used%20to%20extract%20the%20agent%20policy.%20We%20study%20our%20approach%20in%20the%20mobile%20device%20control%20setting.%20Digi-Q%20trains%20the%20Q-function%20using%20offline%20temporal-difference%20%28TD%29%20learning%2C%20on%20top%20of%20frozen%2C%20intermediate-layer%20features%20of%20a%20VLM.%20Compared%20to%20fine-tuning%20the%20whole%20VLM%2C%20this%20approach%20saves%20us%20compute%20and%20enhances%20scalability.%20To%20make%20the%20VLM%20features%20amenable%20for%20representing%20the%20Q-function%2C%20we%20need%20to%20employ%20an%20initial%20phase%20of%20fine-tuning%20to%20amplify%20coverage%20over%20actionable%20information%20needed%20for%20value%20function.%20Once%20trained%2C%20we%20use%20this%20Q-function%20via%20a%20Best-of-N%20policy%20extraction%20operator%20that%20imitates%20the%20best%20action%20out%20of%20multiple%20candidate%20actions%20from%20the%20current%20policy%20as%20ranked%20by%20the%20value%20function%2C%20enabling%20policy%20improvement%20without%20environment%20interaction.%20Digi-Q%20outperforms%20several%20prior%20methods%20on%20user-scale%20device%20control%20tasks%20in%20Android-in-the-Wild%2C%20attaining%2021.2%25%20improvement%20over%20prior%20best-performing%20method.%20In%20some%20cases%2C%20our%20Digi-Q%20approach%20already%20matches%20state-of-the-art%20RL%20methods%20that%20require%20interaction.%20The%20project%20is%20open-sourced%20at%20https%3A%5C%2F%5C%2Fgithub.com%5C%2FDigiRL-agent%5C%2Fdigiq%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22arXiv%3A2502.15760%22%2C%22date%22%3A%222025-02-13%22%2C%22DOI%22%3A%2210.48550%5C%2FarXiv.2502.15760%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22http%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2502.15760%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222025-03-06T21%3A33%3A43Z%22%7D%7D%2C%7B%22key%22%3A%22M756D2UD%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Xu%20et%20al.%22%2C%22parsedDate%22%3A%222024-10-11%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%3Cdiv%20class%3D%5C%22csl-bib-body%5C%22%20style%3D%5C%22line-height%3A%202%3B%20%5C%22%3E%5Cn%20%20%3Cdiv%20class%3D%5C%22csl-entry%5C%22%20style%3D%5C%22clear%3A%20left%3B%20%5C%22%3E%5Cn%20%20%20%20%3Cdiv%20class%3D%5C%22csl-left-margin%5C%22%20style%3D%5C%22float%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%5C%22%3E1.%3C%5C%2Fdiv%3E%3Cdiv%20class%3D%5C%22csl-right-inline%5C%22%20style%3D%5C%22margin%3A%200%20.4em%200%201.5em%3B%5C%22%3EXu%2C%20Z.%2C%20Gupta%2C%20A.%2C%20Li%2C%20T.%2C%20Bentham%2C%20O.%20%26amp%3B%20Srikumar%2C%20V.%20Beyond%20Perplexity%3A%20Multi-dimensional%20Safety%20Evaluation%20of%20LLM%20Compression.%20Preprint%20at%20%3Ca%20class%3D%27zp-DOIURL%27%20href%3D%27https%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FarXiv.2407.04965%27%3Ehttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FarXiv.2407.04965%3C%5C%2Fa%3E%20%282024%29.%3C%5C%2Fdiv%3E%5Cn%20%20%3C%5C%2Fdiv%3E%5Cn%3C%5C%2Fdiv%3E%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22Beyond%20Perplexity%3A%20Multi-dimensional%20Safety%20Evaluation%20of%20LLM%20Compression%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Zhichao%22%2C%22lastName%22%3A%22Xu%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Ashim%22%2C%22lastName%22%3A%22Gupta%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Tao%22%2C%22lastName%22%3A%22Li%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Oliver%22%2C%22lastName%22%3A%22Bentham%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Vivek%22%2C%22lastName%22%3A%22Srikumar%22%7D%5D%2C%22abstractNote%22%3A%22Increasingly%2C%20model%20compression%20techniques%20enable%20large%20language%20models%20%28LLMs%29%20to%20be%20deployed%20in%20real-world%20applications.%20As%20a%20result%20of%20this%20momentum%20towards%20local%20deployment%2C%20compressed%20LLMs%20will%20interact%20with%20a%20large%20population.%20Prior%20work%20on%20compression%20typically%20prioritize%20preserving%20perplexity%2C%20which%20is%20directly%20analogous%20to%20training%20loss.%20The%20impact%20of%20compression%20method%20on%20other%20critical%20aspects%20of%20model%20behavior%5C%5C%2C%20--%20%5C%5C%2Cparticularly%20safety%5C%5C%2C%20--%20%5C%5C%2Crequires%20systematic%20assessment.%20To%20this%20end%2C%20we%20investigate%20the%20impact%20of%20model%20compression%20along%20four%20dimensions%3A%20%281%29%20degeneration%20harm%2C%20i.e.%2C%20bias%20and%20toxicity%20in%20generation%3B%20%282%29%20representational%20harm%2C%20i.e.%2C%20biases%20in%20discriminative%20tasks%3B%20%283%29%20dialect%20bias%3B%20and%284%29%20language%20modeling%20and%20downstream%20task%20performance.%20We%20examine%20a%20wide%20spectrum%20of%20LLM%20compression%20techniques%2C%20including%20unstructured%20pruning%2C%20semi-structured%20pruning%2C%20and%20quantization.%20Our%20analysis%20reveals%20that%20compression%20can%20lead%20to%20unexpected%20consequences.%20Although%20compression%20may%20unintentionally%20alleviate%20LLMs%27%20degeneration%20harm%2C%20it%20can%20still%20exacerbate%20representational%20harm.%20Furthermore%2C%20increasing%20compression%20produces%20a%20divergent%20impact%20on%20different%20protected%20groups.%20Finally%2C%20different%20compression%20methods%20have%20drastically%20different%20safety%20impacts%3A%20for%20example%2C%20quantization%20mostly%20preserves%20bias%20while%20pruning%20degrades%20quickly.%20Our%20findings%20underscore%20the%20importance%20of%20integrating%20safety%20assessments%20into%20the%20development%20of%20compressed%20LLMs%20to%20ensure%20their%20reliability%20across%20real-world%20applications.%5C%5Cfootnote%7BOur%20implementation%20and%20results%20are%20available%20here%3A%20%5C%5Curl%7Bhttps%3A%5C%2F%5C%2Fgithub.com%5C%2Fzhichaoxu-shufe%5C%2FBeyond-Perplexity-Compression-Safety-Eval%7D%7D%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22arXiv%3A2407.04965%22%2C%22date%22%3A%222024-10-11%22%2C%22DOI%22%3A%2210.48550%5C%2FarXiv.2407.04965%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22http%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2407.04965%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222025-03-06T21%3A23%3A00Z%22%7D%7D%2C%7B%22key%22%3A%22KXAUA9TF%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Xu%20et%20al.%22%2C%22parsedDate%22%3A%222025%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%3Cdiv%20class%3D%5C%22csl-bib-body%5C%22%20style%3D%5C%22line-height%3A%202%3B%20%5C%22%3E%5Cn%20%20%3Cdiv%20class%3D%5C%22csl-entry%5C%22%20style%3D%5C%22clear%3A%20left%3B%20%5C%22%3E%5Cn%20%20%20%20%3Cdiv%20class%3D%5C%22csl-left-margin%5C%22%20style%3D%5C%22float%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%5C%22%3E1.%3C%5C%2Fdiv%3E%3Cdiv%20class%3D%5C%22csl-right-inline%5C%22%20style%3D%5C%22margin%3A%200%20.4em%200%201.5em%3B%5C%22%3EXu%2C%20Z.%20%3Ci%3Eet%20al.%3C%5C%2Fi%3E%20A%20Survey%20of%20Model%20Architectures%20in%20Information%20Retrieval.%20Preprint%20at%20%3Ca%20class%3D%27zp-DOIURL%27%20href%3D%27https%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2502.14822%27%3Ehttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2502.14822%3C%5C%2Fa%3E%20%282025%29.%3C%5C%2Fdiv%3E%5Cn%20%20%3C%5C%2Fdiv%3E%5Cn%3C%5C%2Fdiv%3E%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22A%20Survey%20of%20Model%20Architectures%20in%20Information%20Retrieval%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Zhichao%22%2C%22lastName%22%3A%22Xu%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Fengran%22%2C%22lastName%22%3A%22Mo%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Zhiqi%22%2C%22lastName%22%3A%22Huang%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Crystina%22%2C%22lastName%22%3A%22Zhang%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Puxuan%22%2C%22lastName%22%3A%22Yu%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Bei%22%2C%22lastName%22%3A%22Wang%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Jimmy%22%2C%22lastName%22%3A%22Lin%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Vivek%22%2C%22lastName%22%3A%22Srikumar%22%7D%5D%2C%22abstractNote%22%3A%22This%20survey%20examines%20the%20evolution%20of%20model%20architectures%20in%20information%20retrieval%20%28IR%29%2C%20focusing%20on%20two%20key%20aspects%3A%20backbone%20models%20for%20feature%20extraction%20and%20end-to-end%20system%20architectures%20for%20relevance%20estimation.%20The%20review%20intentionally%20separates%20architectural%20considerations%20from%20training%20methodologies%20to%20provide%20a%20focused%20analysis%20of%20structural%20innovations%20in%20IR%20systems.We%20trace%20the%20development%20from%20traditional%20term-based%20methods%20to%20modern%20neural%20approaches%2C%20particularly%20highlighting%20the%20impact%20of%20transformer-based%20models%20and%20subsequent%20large%20language%20models%20%28LLMs%29.%20We%20conclude%20by%20discussing%20emerging%20challenges%20and%20future%20directions%2C%20including%20architectural%20optimizations%20for%20performance%20and%20scalability%2C%20handling%20of%20multimodal%2C%20multilingual%20data%2C%20and%20adaptation%20to%20novel%20application%20domains%20beyond%20traditional%20search%20paradigms.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222025%22%2C%22DOI%22%3A%2210.48550%5C%2FARXIV.2502.14822%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2502.14822%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222025-03-06T21%3A20%3A12Z%22%7D%7D%2C%7B%22key%22%3A%22KLI6A85N%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Li%20and%20Luo%22%2C%22parsedDate%22%3A%222025-02-18%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%3Cdiv%20class%3D%5C%22csl-bib-body%5C%22%20style%3D%5C%22line-height%3A%202%3B%20%5C%22%3E%5Cn%20%20%3Cdiv%20class%3D%5C%22csl-entry%5C%22%20style%3D%5C%22clear%3A%20left%3B%20%5C%22%3E%5Cn%20%20%20%20%3Cdiv%20class%3D%5C%22csl-left-margin%5C%22%20style%3D%5C%22float%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%5C%22%3E1.%3C%5C%2Fdiv%3E%3Cdiv%20class%3D%5C%22csl-right-inline%5C%22%20style%3D%5C%22margin%3A%200%20.4em%200%201.5em%3B%5C%22%3ELi%2C%20Z.%20%26amp%3B%20Luo%2C%20Y.%20Rewiring%20protein%20sequence%20and%20structure%20generative%20models%20to%20enhance%20protein%20stability%20prediction.%20Preprint%20at%20%3Ca%20class%3D%27zp-DOIURL%27%20href%3D%27https%3A%5C%2F%5C%2Fdoi.org%5C%2F10.1101%5C%2F2025.02.13.638154%27%3Ehttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.1101%5C%2F2025.02.13.638154%3C%5C%2Fa%3E%20%282025%29.%3C%5C%2Fdiv%3E%5Cn%20%20%3C%5C%2Fdiv%3E%5Cn%3C%5C%2Fdiv%3E%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22Rewiring%20protein%20sequence%20and%20structure%20generative%20models%20to%20enhance%20protein%20stability%20prediction%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Ziang%22%2C%22lastName%22%3A%22Li%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Yunan%22%2C%22lastName%22%3A%22Luo%22%7D%5D%2C%22abstractNote%22%3A%22Abstract%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20Predicting%20changes%20in%20protein%20thermostability%20due%20to%20amino%20acid%20substitutions%20is%20essential%20for%20understanding%20human%20diseases%20and%20engineering%20useful%20proteins%20for%20clinical%20and%20industrial%20applications.%20While%20recent%20advances%20in%20protein%20generative%20models%2C%20which%20learn%20probability%20distributions%20over%20amino%20acids%20conditioned%20on%20structural%20or%20evolutionary%20sequence%20contexts%2C%20have%20shown%20impressive%20performance%20in%20predicting%20various%20protein%20properties%20without%20task-specific%20training%2C%20their%20strong%20unsupervised%20prediction%20ability%20does%20not%20extend%20to%20all%20protein%20functions.%20In%20particular%2C%20their%20potential%20to%20improve%20protein%20stability%20prediction%20remains%20underexplored.%20In%20this%20work%2C%20we%20present%20SPURS%2C%20a%20novel%20deep%20learning%20framework%20that%20adapts%20and%20integrates%20two%20general-purpose%20protein%20generative%20models%5Cu2013a%20protein%20language%20model%20%28ESM%29%20and%20an%20inverse%20folding%20model%20%28ProteinMPNN%29%5Cu2013into%20an%20effective%20stability%20predictor.%20SPURS%20employs%20a%20lightweight%20neural%20network%20module%20to%20rewire%20per-residue%20structure%20representations%20learned%20by%20ProteinMPNN%20into%20the%20attention%20layers%20of%20ESM%2C%20thereby%20informing%20and%20enhancing%20ESM%5Cu2019s%20sequence%20representation%20learning.%20This%20rewiring%20strategy%20enables%20SPURS%20to%20harness%20evolutionary%20patterns%20from%20both%20sequence%20and%20structure%20data%2C%20where%20the%20sequence%20like-lihood%20distribution%20learned%20by%20ESM%20is%20conditioned%20on%20structure%20priors%20encoded%20by%20ProteinMPNN%20to%20predict%20mutation%20effects.%20We%20steer%20this%20integrated%20framework%20to%20a%20stability%20prediction%20model%20through%20supervised%20training%20on%20a%20recently%20released%20mega-scale%20thermostability%20dataset.%20Evaluations%20across%2012%20benchmark%20datasets%20showed%20that%20SPURS%20delivers%20accurate%2C%20rapid%2C%20scalable%2C%20and%20generalizable%20stability%20predictions%2C%20consistently%20outperforming%20current%20state-of-the-art%20methods.%20Notably%2C%20SPURS%20demonstrates%20remarkable%20versatility%20in%20protein%20stability%20and%20function%20analyses%3A%20when%20combined%20with%20a%20protein%20language%20model%2C%20it%20accurately%20identifies%20protein%20functional%20sites%20in%20an%20unsupervised%20manner.%20Additionally%2C%20it%20enhances%20current%20low-%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20N%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20protein%20fitness%20prediction%20models%20by%20serving%20as%20a%20stability%20prior%20model%20to%20improve%20accuracy.%20These%20results%20highlight%20SPURS%20as%20a%20powerful%20tool%20to%20advance%20current%20protein%20stability%20prediction%20and%20machine%20learning-guided%20protein%20engineering%20workflows.%20The%20source%20code%20of%20SPURS%20is%20available%20at%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20https%3A%5C%2F%5C%2Fgithub.com%5C%2Fluo-group%5C%2FSPURS%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222025-02-18%22%2C%22DOI%22%3A%2210.1101%5C%2F2025.02.13.638154%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22http%3A%5C%2F%5C%2Fbiorxiv.org%5C%2Flookup%5C%2Fdoi%5C%2F10.1101%5C%2F2025.02.13.638154%22%2C%22language%22%3A%22en%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222025-03-06T21%3A14%3A34Z%22%7D%7D%2C%7B%22key%22%3A%22DUG3A9CG%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Gangwar%20et%20al.%22%2C%22parsedDate%22%3A%222025%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%3Cdiv%20class%3D%5C%22csl-bib-body%5C%22%20style%3D%5C%22line-height%3A%202%3B%20%5C%22%3E%5Cn%20%20%3Cdiv%20class%3D%5C%22csl-entry%5C%22%20style%3D%5C%22clear%3A%20left%3B%20%5C%22%3E%5Cn%20%20%20%20%3Cdiv%20class%3D%5C%22csl-left-margin%5C%22%20style%3D%5C%22float%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%5C%22%3E1.%3C%5C%2Fdiv%3E%3Cdiv%20class%3D%5C%22csl-right-inline%5C%22%20style%3D%5C%22margin%3A%200%20.4em%200%201.5em%3B%5C%22%3EGangwar%2C%20N.%2C%20Bhat%2C%20S.%20P.%20%26amp%3B%20Kani%2C%20N.%20Integrating%20Arithmetic%20Learning%20Improves%20Mathematical%20Reasoning%20in%20Smaller%20Models.%20Preprint%20at%20%3Ca%20class%3D%27zp-DOIURL%27%20href%3D%27https%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2502.12855%27%3Ehttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2502.12855%3C%5C%2Fa%3E%20%282025%29.%3C%5C%2Fdiv%3E%5Cn%20%20%3C%5C%2Fdiv%3E%5Cn%3C%5C%2Fdiv%3E%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22Integrating%20Arithmetic%20Learning%20Improves%20Mathematical%20Reasoning%20in%20Smaller%20Models%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Neeraj%22%2C%22lastName%22%3A%22Gangwar%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Suma%20P%22%2C%22lastName%22%3A%22Bhat%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Nickvash%22%2C%22lastName%22%3A%22Kani%22%7D%5D%2C%22abstractNote%22%3A%22While%20large%20models%20pre-trained%20on%20high-quality%20data%20exhibit%20excellent%20performance%20across%20various%20reasoning%20tasks%2C%20including%20mathematical%20reasoning%20%28e.g.%20GSM8k%2C%20MultiArith%29%2C%20specializing%20smaller%20models%20to%20excel%20at%20mathematical%20reasoning%20remains%20a%20challenging%20problem.%20Common%20approaches%20to%20address%20this%20challenge%20include%20knowledge%20distillation%2C%20where%20smaller%20student%20models%20learn%20from%20large%20pre-trained%20teacher%20models%2C%20and%20data%20augmentation%2C%20such%20as%20rephrasing%20questions.%20Despite%20these%20efforts%2C%20smaller%20models%20struggle%20with%20arithmetic%20computations%2C%20leading%20to%20errors%20in%20mathematical%20reasoning.%20In%20this%20work%2C%20we%20focus%20on%20leveraging%20a%20programmatically%20generated%20arithmetic%20dataset%20to%20enhance%20the%20reasoning%20capabilities%20of%20smaller%20models.%20We%20investigate%20two%20key%20approaches%20to%20incorporate%20this%20dataset%20--%20%281%29%20intermediate%20fine-tuning%2C%20where%20a%20model%20is%20fine-tuned%20on%20the%20arithmetic%20dataset%20before%20being%20trained%20on%20a%20reasoning%20dataset%2C%20and%20%282%29%20integrating%20the%20arithmetic%20dataset%20into%20the%20instruction-tuning%20mixture%2C%20allowing%20the%20model%20to%20learn%20arithmetic%20skills%20alongside%20general%20instruction-following%20abilities.%20Our%20experiments%20on%20multiple%20reasoning%20benchmarks%20demonstrate%20that%20incorporating%20an%20arithmetic%20dataset%2C%20whether%20through%20targeted%20fine-tuning%20or%20within%20the%20instruction-tuning%20mixture%2C%20enhances%20the%20models%27%20arithmetic%20capabilities%2C%20which%20in%20turn%20improves%20their%20mathematical%20reasoning%20performance.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222025%22%2C%22DOI%22%3A%2210.48550%5C%2FARXIV.2502.12855%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2502.12855%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222025-03-06T20%3A54%3A19Z%22%7D%7D%2C%7B%22key%22%3A%22KTERPC2I%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Srivastava%20et%20al.%22%2C%22parsedDate%22%3A%222025%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%3Cdiv%20class%3D%5C%22csl-bib-body%5C%22%20style%3D%5C%22line-height%3A%202%3B%20%5C%22%3E%5Cn%20%20%3Cdiv%20class%3D%5C%22csl-entry%5C%22%20style%3D%5C%22clear%3A%20left%3B%20%5C%22%3E%5Cn%20%20%20%20%3Cdiv%20class%3D%5C%22csl-left-margin%5C%22%20style%3D%5C%22float%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%5C%22%3E1.%3C%5C%2Fdiv%3E%3Cdiv%20class%3D%5C%22csl-right-inline%5C%22%20style%3D%5C%22margin%3A%200%20.4em%200%201.5em%3B%5C%22%3ESrivastava%2C%20G.%2C%20Cao%2C%20S.%20%26amp%3B%20Wang%2C%20X.%20Towards%20Reasoning%20Ability%20of%20Small%20Language%20Models.%20Preprint%20at%20%3Ca%20class%3D%27zp-DOIURL%27%20href%3D%27https%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2502.11569%27%3Ehttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2502.11569%3C%5C%2Fa%3E%20%282025%29.%3C%5C%2Fdiv%3E%5Cn%20%20%3C%5C%2Fdiv%3E%5Cn%3C%5C%2Fdiv%3E%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22Towards%20Reasoning%20Ability%20of%20Small%20Language%20Models%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Gaurav%22%2C%22lastName%22%3A%22Srivastava%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Shuxiang%22%2C%22lastName%22%3A%22Cao%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Xuan%22%2C%22lastName%22%3A%22Wang%22%7D%5D%2C%22abstractNote%22%3A%22Reasoning%20has%20long%20been%20viewed%20as%20an%20emergent%20property%20of%20large%20language%20models%20%28LLMs%29%2C%20appearing%20at%20or%20above%20a%20certain%20scale%20%28%24%5C%5Csim%24100B%20parameters%29.%20However%2C%20recent%20studies%20challenge%20this%20assumption%2C%20showing%20that%20small%20language%20models%20%28SLMs%29%20can%20also%20achieve%20competitive%20reasoning%20performance.%20SLMs%20are%20increasingly%20favored%20for%20their%20efficiency%20and%20deployability.%20However%2C%20there%20is%20a%20lack%20of%20systematic%20study%20on%20the%20reasoning%20abilities%20of%20diverse%20SLMs%2C%20including%20those%20trained%20from%20scratch%20or%20derived%20from%20LLMs%20through%20quantization%2C%20pruning%2C%20and%20distillation.%20This%20raises%20a%20critical%20question%3A%20Can%20SLMs%20achieve%20reasoning%20abilities%20comparable%20to%20LLMs%3F%20In%20this%20work%2C%20we%20systematically%20survey%2C%20benchmark%2C%20and%20analyze%2072%20SLMs%20from%20six%20model%20families%20across%2014%20reasoning%20benchmarks.%20For%20reliable%20evaluation%2C%20we%20examine%20four%20evaluation%20methods%20and%20compare%20four%20LLM%20judges%20against%20human%20evaluations%20on%20800%20data%20points.%20We%20repeat%20all%20experiments%20three%20times%20to%20ensure%20a%20robust%20performance%20assessment.%20Additionally%2C%20we%20analyze%20the%20impact%20of%20different%20prompting%20strategies%20in%20small%20models.%20Beyond%20accuracy%2C%20we%20also%20evaluate%20model%20robustness%20under%20adversarial%20conditions%20and%20intermediate%20reasoning%20steps.%20Our%20findings%20challenge%20the%20assumption%20that%20scaling%20is%20the%20only%20way%20to%20achieve%20strong%20reasoning.%20Instead%2C%20we%20foresee%20a%20future%20where%20SLMs%20with%20strong%20reasoning%20capabilities%20can%20be%20developed%20through%20structured%20training%20or%20post-training%20compression.%20They%20can%20serve%20as%20efficient%20alternatives%20to%20LLMs%20for%20reasoning-intensive%20tasks.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222025%22%2C%22DOI%22%3A%2210.48550%5C%2FARXIV.2502.11569%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2502.11569%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222025-03-06T20%3A51%3A57Z%22%7D%7D%2C%7B%22key%22%3A%22JX55A7QN%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Luo%20and%20Luo%22%2C%22parsedDate%22%3A%222025-02-17%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%3Cdiv%20class%3D%5C%22csl-bib-body%5C%22%20style%3D%5C%22line-height%3A%202%3B%20%5C%22%3E%5Cn%20%20%3Cdiv%20class%3D%5C%22csl-entry%5C%22%20style%3D%5C%22clear%3A%20left%3B%20%5C%22%3E%5Cn%20%20%20%20%3Cdiv%20class%3D%5C%22csl-left-margin%5C%22%20style%3D%5C%22float%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%5C%22%3E1.%3C%5C%2Fdiv%3E%3Cdiv%20class%3D%5C%22csl-right-inline%5C%22%20style%3D%5C%22margin%3A%200%20.4em%200%201.5em%3B%5C%22%3ELuo%2C%20J.%20%26amp%3B%20Luo%2C%20Y.%20Learning%20maximally%20spanning%20representations%20improves%20protein%20function%20annotation.%20Preprint%20at%20%3Ca%20class%3D%27zp-DOIURL%27%20href%3D%27https%3A%5C%2F%5C%2Fdoi.org%5C%2F10.1101%5C%2F2025.02.13.638156%27%3Ehttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.1101%5C%2F2025.02.13.638156%3C%5C%2Fa%3E%20%282025%29.%3C%5C%2Fdiv%3E%5Cn%20%20%3C%5C%2Fdiv%3E%5Cn%3C%5C%2Fdiv%3E%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22Learning%20maximally%20spanning%20representations%20improves%20protein%20function%20annotation%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Jiaqi%22%2C%22lastName%22%3A%22Luo%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Yunan%22%2C%22lastName%22%3A%22Luo%22%7D%5D%2C%22abstractNote%22%3A%22Abstract%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20Automated%20protein%20function%20annotation%20is%20a%20fundamental%20problem%20in%20computational%20biology%2C%20crucial%20for%20understanding%20the%20functional%20roles%20of%20proteins%20in%20biological%20processes%2C%20with%20broad%20implications%20in%20medicine%20and%20biotechnology.%20A%20persistent%20challenge%20in%20this%20problem%20is%20the%20imbalanced%2C%20long-tail%20distribution%20of%20available%20function%20annotations%3A%20a%20small%20set%20of%20well-studied%20function%20classes%20account%20for%20most%20annotated%20proteins%2C%20while%20many%20other%20classes%20have%20few%20annotated%20proteins%2C%20often%20due%20to%20investigative%20bias%2C%20experimental%20limitations%2C%20or%20intrinsic%20biases%20in%20protein%20evolution.%20As%20a%20result%2C%20existing%20machine%20learning%20models%20for%20protein%20function%20prediction%20tend%20to%20only%20optimize%20the%20prediction%20accuracy%20for%20well-studied%20function%20classes%20overrepresented%20in%20the%20training%20data%2C%20leading%20to%20poor%20accuracy%20for%20understudied%20functions.%20In%20this%20work%2C%20we%20develop%20MSRep%2C%20a%20novel%20deep%20learning-based%20protein%20function%20annotation%20framework%20designed%20to%20address%20this%20imbalance%20issue%20and%20improve%20annotation%20accuracy.%20MSRep%20is%20inspired%20by%20an%20intriguing%20phenomenon%2C%20called%20neural%20collapse%20%28NC%29%2C%20commonly%20observed%20in%20high-accuracy%20deep%20neural%20networks%20used%20for%20classification%20tasks%2C%20where%20hidden%20representations%20in%20the%20final%20layer%20collapse%20to%20class-specific%20mean%20embeddings%2C%20while%20maintaining%20maximal%20inter-class%20separation.%20Given%20that%20NC%20consistently%20emerges%20across%20diverse%20architectures%20and%20tasks%20for%20high-accuracy%20models%2C%20we%20hypothesize%20that%20inducing%20NC%20structure%20in%20models%20trained%20on%20imbalanced%20data%20can%20enhance%20both%20prediction%20accuracy%20and%20generalizability.%20To%20achieve%20this%2C%20MSRep%20refines%20a%20pre-trained%20protein%20language%20model%20to%20produce%20NC-like%20representations%20by%20optimizing%20an%20NC-inspired%20loss%20function%2C%20which%20ensures%20that%20minority%20functions%20are%20equally%20represented%20in%20the%20embedding%20space%20as%20majority%20functions%2C%20in%20contrast%20to%20conventional%20classification%20methods%20whose%20embedding%20spaces%20are%20dominated%20by%20overrepresented%20classes.%20In%20evaluations%20across%20four%20protein%20function%20annotation%20tasks%20on%20the%20prediction%20of%20Enzyme%20Commission%20numbers%2C%20Gene3D%20codes%2C%20Pfam%20families%2C%20and%20Gene%20Ontology%20terms%2C%20MSRep%20demonstrates%20superior%20predictive%20performance%20for%20both%20well-%20and%20underrepresented%20classes%2C%20outperforming%20several%20state-of-the-art%20annotation%20tools.%20We%20anticipate%20that%20MSRep%20will%20enhance%20the%20annotation%20of%20understudied%20functions%20and%20novel%2C%20uncharacterized%20proteins%2C%20advancing%20future%20protein%20function%20studies%20and%20accelerating%20the%20discovery%20of%20new%20functional%20proteins.%20The%20source%20code%20of%20MSRep%20is%20available%20at%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20https%3A%5C%2F%5C%2Fgithub.com%5C%2Fluo-group%5C%2FMSRep%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222025-02-17%22%2C%22DOI%22%3A%2210.1101%5C%2F2025.02.13.638156%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22http%3A%5C%2F%5C%2Fbiorxiv.org%5C%2Flookup%5C%2Fdoi%5C%2F10.1101%5C%2F2025.02.13.638156%22%2C%22language%22%3A%22en%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222025-03-06T20%3A22%3A29Z%22%7D%7D%2C%7B%22key%22%3A%222DGY3BJ8%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Chen%20et%20al.%22%2C%22parsedDate%22%3A%222025%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%3Cdiv%20class%3D%5C%22csl-bib-body%5C%22%20style%3D%5C%22line-height%3A%202%3B%20%5C%22%3E%5Cn%20%20%3Cdiv%20class%3D%5C%22csl-entry%5C%22%20style%3D%5C%22clear%3A%20left%3B%20%5C%22%3E%5Cn%20%20%20%20%3Cdiv%20class%3D%5C%22csl-left-margin%5C%22%20style%3D%5C%22float%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%5C%22%3E1.%3C%5C%2Fdiv%3E%3Cdiv%20class%3D%5C%22csl-right-inline%5C%22%20style%3D%5C%22margin%3A%200%20.4em%200%201.5em%3B%5C%22%3EChen%2C%20W.%20%3Ci%3Eet%20al.%3C%5C%2Fi%3E%20OWLS%3A%20Scaling%20Laws%20for%20Multilingual%20Speech%20Recognition%20and%20Translation%20Models.%20Preprint%20at%20%3Ca%20class%3D%27zp-DOIURL%27%20href%3D%27https%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2502.10373%27%3Ehttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2502.10373%3C%5C%2Fa%3E%20%282025%29.%3C%5C%2Fdiv%3E%5Cn%20%20%3C%5C%2Fdiv%3E%5Cn%3C%5C%2Fdiv%3E%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22OWLS%3A%20Scaling%20Laws%20for%20Multilingual%20Speech%20Recognition%20and%20Translation%20Models%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22William%22%2C%22lastName%22%3A%22Chen%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Jinchuan%22%2C%22lastName%22%3A%22Tian%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Yifan%22%2C%22lastName%22%3A%22Peng%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Brian%22%2C%22lastName%22%3A%22Yan%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Chao-Han%20Huck%22%2C%22lastName%22%3A%22Yang%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Shinji%22%2C%22lastName%22%3A%22Watanabe%22%7D%5D%2C%22abstractNote%22%3A%22Neural%20scaling%20laws%20offer%20valuable%20insights%20for%20designing%20robust%20sequence%20processing%20architectures.%20While%20these%20laws%20have%20been%20extensively%20characterized%20in%20other%20modalities%2C%20their%20behavior%20in%20speech%20remains%20comparatively%20underexplored.%20In%20this%20work%2C%20we%20introduce%20OWLS%2C%20an%20open-access%2C%20reproducible%20suite%20of%20multilingual%20speech%20recognition%20and%20translation%20models%20spanning%200.25B%20to%2018B%20parameters%2C%20with%20the%2018B%20version%20being%20the%20largest%20speech%20model%2C%20to%20the%20best%20of%20our%20knowledge.%20OWLS%20leverages%20up%20to%20360K%20hours%20of%20public%20speech%20data%20across%20150%20languages%2C%20enabling%20a%20systematic%20investigation%20into%20how%20data%2C%20model%2C%20and%20compute%20scaling%20each%20influence%20performance%20in%20multilingual%20speech%20tasks.%20We%20use%20OWLS%20to%20derive%20neural%20scaling%20laws%2C%20showing%20how%20final%20performance%20can%20be%20reliably%20predicted%20when%20scaling.%20One%20of%20our%20key%20findings%20is%20that%20scaling%20enhances%20performance%20on%20low-resource%20languages%5C%2Fdialects%2C%20helping%20to%20mitigate%20bias%20and%20improve%20the%20accessibility%20of%20speech%20technologies.%20Finally%2C%20we%20show%20how%20OWLS%20can%20be%20used%20to%20power%20new%20research%20directions%20by%20discovering%20emergent%20abilities%20in%20large-scale%20speech%20models.%20Model%20checkpoints%20will%20be%20released%20on%20https%3A%5C%2F%5C%2Fhuggingface.co%5C%2Fcollections%5C%2Fespnet%5C%2Fowls-scaling-laws-for-speech-recognition-and-translation-67ab7f991c194065f057ce8d%20for%20future%20studies.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222025%22%2C%22DOI%22%3A%2210.48550%5C%2FARXIV.2502.10373%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2502.10373%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222025-03-06T20%3A19%3A44Z%22%7D%7D%2C%7B%22key%22%3A%22CX7E22T2%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Ozgulbas%20et%20al.%22%2C%22parsedDate%22%3A%222025-02-14%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%3Cdiv%20class%3D%5C%22csl-bib-body%5C%22%20style%3D%5C%22line-height%3A%202%3B%20%5C%22%3E%5Cn%20%20%3Cdiv%20class%3D%5C%22csl-entry%5C%22%20style%3D%5C%22clear%3A%20left%3B%20%5C%22%3E%5Cn%20%20%20%20%3Cdiv%20class%3D%5C%22csl-left-margin%5C%22%20style%3D%5C%22float%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%5C%22%3E1.%3C%5C%2Fdiv%3E%3Cdiv%20class%3D%5C%22csl-right-inline%5C%22%20style%3D%5C%22margin%3A%200%20.4em%200%201.5em%3B%5C%22%3EOzgulbas%2C%20D.%20G.%20%3Ci%3Eet%20al.%3C%5C%2Fi%3E%20Probing%20the%20Role%20of%20Membrane%20in%20Neutralizing%20Activity%20of%20Antibodies%20Against%20Influenza%20Virus.%20Preprint%20at%20%3Ca%20class%3D%27zp-DOIURL%27%20href%3D%27https%3A%5C%2F%5C%2Fdoi.org%5C%2F10.1101%5C%2F2025.02.11.637756%27%3Ehttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.1101%5C%2F2025.02.11.637756%3C%5C%2Fa%3E%20%282025%29.%3C%5C%2Fdiv%3E%5Cn%20%20%3C%5C%2Fdiv%3E%5Cn%3C%5C%2Fdiv%3E%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22Probing%20the%20Role%20of%20Membrane%20in%20Neutralizing%20Activity%20of%20Antibodies%20Against%20Influenza%20Virus%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Defne%20G.%22%2C%22lastName%22%3A%22Ozgulbas%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Timothy%20J.%20C.%22%2C%22lastName%22%3A%22Tan%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Po-Chao%22%2C%22lastName%22%3A%22Wen%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Qi%20Wen%22%2C%22lastName%22%3A%22Teo%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Huibin%22%2C%22lastName%22%3A%22Lv%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Zhaleh%22%2C%22lastName%22%3A%22Ghaemi%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Martin%22%2C%22lastName%22%3A%22Frank%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Nicholas%20C.%22%2C%22lastName%22%3A%22Wu%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Emad%22%2C%22lastName%22%3A%22Tajkhorshid%22%7D%5D%2C%22abstractNote%22%3A%22SUMMARY%20%5Cn%20%20%20%20%20%20%20%20%20%20Influenza%20poses%20a%20major%20health%20issue%20globally.%20Neutralizing%20antibodies%20targeting%20the%20highly%20conserved%20stem%20region%20of%20hemagglutinin%20%28HA%29%20of%20the%20influenza%20virus%20provide%20considerable%20protection%20against%20the%20infection.%20Using%20an%20array%20of%20advanced%20simulation%20technologies%2C%20we%20developed%20a%20high-resolution%20structural%20model%20of%20full-length%2C%20Fab-bound%20HA%20in%20a%20native%20viral%20membrane%20to%20characterize%20direct%20membrane%20interactions%20that%20govern%20the%20efficacy%20of%20the%20antibody.%20We%20reveal%20functionally%20important%20residues%20beyond%20the%20antibody%5Cu2019s%20complementary-determining%20regions%20that%20contribute%20to%20its%20membrane%20binding.%20Mutagenesis%20experiments%20and%20infectivity%20assays%20confirm%20that%20deactivating%20the%20membrane-binding%20residues%20of%20the%20antibody%20decreases%20its%20neutralization%20activity.%20Therefore%2C%20we%20propose%20that%20the%20association%20with%20the%20viral%20membrane%20plays%20a%20key%20role%20in%20the%20neutralization%20activity%20of%20these%20antibodies.%20Given%20the%20rapid%20evolution%20of%20the%20influenza%20virus%2C%20the%20developed%20model%20provides%20a%20structural%20framework%20for%20the%20rational%20design%20and%20development%20of%20more%20effective%20therapeutic%20antibodies.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222025-02-14%22%2C%22DOI%22%3A%2210.1101%5C%2F2025.02.11.637756%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22http%3A%5C%2F%5C%2Fbiorxiv.org%5C%2Flookup%5C%2Fdoi%5C%2F10.1101%5C%2F2025.02.11.637756%22%2C%22language%22%3A%22en%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222025-03-06T19%3A23%3A21Z%22%7D%7D%2C%7B%22key%22%3A%22RI3VS4UF%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Abedsoltan%20et%20al.%22%2C%22parsedDate%22%3A%222025%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%3Cdiv%20class%3D%5C%22csl-bib-body%5C%22%20style%3D%5C%22line-height%3A%202%3B%20%5C%22%3E%5Cn%20%20%3Cdiv%20class%3D%5C%22csl-entry%5C%22%20style%3D%5C%22clear%3A%20left%3B%20%5C%22%3E%5Cn%20%20%20%20%3Cdiv%20class%3D%5C%22csl-left-margin%5C%22%20style%3D%5C%22float%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%5C%22%3E1.%3C%5C%2Fdiv%3E%3Cdiv%20class%3D%5C%22csl-right-inline%5C%22%20style%3D%5C%22margin%3A%200%20.4em%200%201.5em%3B%5C%22%3EAbedsoltan%2C%20A.%20%3Ci%3Eet%20al.%3C%5C%2Fi%3E%20Task%20Generalization%20With%20AutoRegressive%20Compositional%20Structure%3A%20Can%20Learning%20From%20%24%5C%5Cd%24%20Tasks%20Generalize%20to%20%24%5C%5Cd%5E%7BT%7D%24%20Tasks%3F%20Preprint%20at%20%3Ca%20class%3D%27zp-DOIURL%27%20href%3D%27https%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2502.08991%27%3Ehttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2502.08991%3C%5C%2Fa%3E%20%282025%29.%3C%5C%2Fdiv%3E%5Cn%20%20%3C%5C%2Fdiv%3E%5Cn%3C%5C%2Fdiv%3E%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22Task%20Generalization%20With%20AutoRegressive%20Compositional%20Structure%3A%20Can%20Learning%20From%20%24%5C%5Cd%24%20Tasks%20Generalize%20to%20%24%5C%5Cd%5E%7BT%7D%24%20Tasks%3F%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Amirhesam%22%2C%22lastName%22%3A%22Abedsoltan%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Huaqing%22%2C%22lastName%22%3A%22Zhang%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Kaiyue%22%2C%22lastName%22%3A%22Wen%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Hongzhou%22%2C%22lastName%22%3A%22Lin%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Jingzhao%22%2C%22lastName%22%3A%22Zhang%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Mikhail%22%2C%22lastName%22%3A%22Belkin%22%7D%5D%2C%22abstractNote%22%3A%22Large%20language%20models%20%28LLMs%29%20exhibit%20remarkable%20task%20generalization%2C%20solving%20tasks%20they%20were%20never%20explicitly%20trained%20on%20with%20only%20a%20few%20demonstrations.%20This%20raises%20a%20fundamental%20question%3A%20When%20can%20learning%20from%20a%20small%20set%20of%20tasks%20generalize%20to%20a%20large%20task%20family%3F%20In%20this%20paper%2C%20we%20investigate%20task%20generalization%20through%20the%20lens%20of%20AutoRegressive%20Compositional%20%28ARC%29%20structure%2C%20where%20each%20task%20is%20a%20composition%20of%20%24T%24%20operations%2C%20and%20each%20operation%20is%20among%20a%20finite%20family%20of%20%24%5C%5Cd%24%20subtasks.%20This%20yields%20a%20total%20class%20of%20size~%5C%5C%28%20%5C%5Cd%5E%5C%5CTT%20%5C%5C%29.%20We%20first%20show%20that%20generalization%20to%20all%20%5C%5C%28%20%5C%5Cd%5E%5C%5CTT%20%5C%5C%29%20tasks%20is%20theoretically%20achievable%20by%20training%20on%20only%20%5C%5C%28%20%5C%5Ctilde%7BO%7D%28%5C%5Cd%29%20%5C%5C%29%20tasks.%20Empirically%2C%20we%20demonstrate%20that%20Transformers%20achieve%20such%20exponential%20task%20generalization%20on%20sparse%20parity%20functions%20via%20in-context%20learning%20%28ICL%29%20and%20Chain-of-Thought%20%28CoT%29%20reasoning.%20We%20further%20demonstrate%20this%20generalization%20in%20arithmetic%20and%20language%20translation%2C%20extending%20beyond%20parity%20functions.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222025%22%2C%22DOI%22%3A%2210.48550%5C%2FARXIV.2502.08991%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2502.08991%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222025-03-06T19%3A06%3A54Z%22%7D%7D%2C%7B%22key%22%3A%2289AD3F7D%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Chu%20et%20al.%22%2C%22parsedDate%22%3A%222024%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%3Cdiv%20class%3D%5C%22csl-bib-body%5C%22%20style%3D%5C%22line-height%3A%202%3B%20%5C%22%3E%5Cn%20%20%3Cdiv%20class%3D%5C%22csl-entry%5C%22%20style%3D%5C%22clear%3A%20left%3B%20%5C%22%3E%5Cn%20%20%20%20%3Cdiv%20class%3D%5C%22csl-left-margin%5C%22%20style%3D%5C%22float%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%5C%22%3E1.%3C%5C%2Fdiv%3E%3Cdiv%20class%3D%5C%22csl-right-inline%5C%22%20style%3D%5C%22margin%3A%200%20.4em%200%201.5em%3B%5C%22%3EChu%2C%20T.%2C%20Estrada%2C%20J.%20B.%20%26amp%3B%20Bryngelson%2C%20S.%20H.%20Bayesian%20optimal%20design%20accelerates%20discovery%20of%20material%20properties%20from%20bubble%20dynamics.%20Preprint%20at%20%3Ca%20class%3D%27zp-DOIURL%27%20href%3D%27https%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2409.00011%27%3Ehttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2409.00011%3C%5C%2Fa%3E%20%282024%29.%3C%5C%2Fdiv%3E%5Cn%20%20%3C%5C%2Fdiv%3E%5Cn%3C%5C%2Fdiv%3E%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22Bayesian%20optimal%20design%20accelerates%20discovery%20of%20material%20properties%20from%20bubble%20dynamics%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Tianyi%22%2C%22lastName%22%3A%22Chu%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Jonathan%20B.%22%2C%22lastName%22%3A%22Estrada%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Spencer%20H.%22%2C%22lastName%22%3A%22Bryngelson%22%7D%5D%2C%22abstractNote%22%3A%22An%20optimal%20sequential%20experimental%20design%20approach%20is%20developed%20to%20computationally%20characterize%20soft%20material%20properties%20at%20the%20high%20strain%20rates%20associated%20with%20bubble%20cavitation.%20The%20approach%20involves%20optimal%20design%20and%20model%20inference.%20The%20optimal%20design%20strategy%20maximizes%20the%20expected%20information%20gain%20in%20a%20Bayesian%20statistical%20setting%20to%20design%20experiments%20that%20provide%20the%20most%20informative%20cavitation%20data%20about%20unknown%20soft%20material%20properties.%20We%20infer%20constitutive%20models%20by%20characterizing%20the%20associated%20viscoelastic%20properties%20from%20measurements%20via%20a%20hybrid%20ensemble-based%204D-Var%20method%20%28En4D-Var%29.%20The%20inertial%20microcavitation-based%20high%20strain-rate%20rheometry%20%28IMR%29%20method%20%28%5B1%5D%29%20simulates%20the%20bubble%20dynamics%20under%20laser-induced%20cavitation.%20We%20use%20experimental%20measurements%20to%20create%20synthetic%20data%20representing%20the%20viscoelastic%20behavior%20of%20stiff%20and%20soft%20polyacrylamide%20hydrogels%20under%20realistic%20uncertainties.%20The%20synthetic%20data%20are%20seeded%20with%20larger%20errors%20than%20state-of-the-art%20measurements%20yet%20match%20known%20material%20properties%2C%20reaching%201%25%20relative%20error%20within%2010%20sequential%20designs%20%28experiments%29.%20We%20discern%20between%20two%20seemingly%20equally%20plausible%20constitutive%20models%2C%20Neo-Hookean%20Kelvin--Voigt%20and%20quadratic%20Kelvin--Voigt%2C%20with%20a%20probability%20of%20correctness%20larger%20than%2099%25%20in%20the%20same%20number%20of%20experiments.%20This%20strategy%20discovers%20soft%20material%20properties%2C%20including%20discriminating%20between%20constitutive%20models%20and%20discerning%20their%20parameters%2C%20using%20only%20a%20few%20experiments.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222024%22%2C%22DOI%22%3A%2210.48550%5C%2FARXIV.2409.00011%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2409.00011%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222025-03-06T19%3A02%3A37Z%22%7D%7D%2C%7B%22key%22%3A%228KLMCI5C%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Choi%20et%20al.%22%2C%22parsedDate%22%3A%222025%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%3Cdiv%20class%3D%5C%22csl-bib-body%5C%22%20style%3D%5C%22line-height%3A%202%3B%20%5C%22%3E%5Cn%20%20%3Cdiv%20class%3D%5C%22csl-entry%5C%22%20style%3D%5C%22clear%3A%20left%3B%20%5C%22%3E%5Cn%20%20%20%20%3Cdiv%20class%3D%5C%22csl-left-margin%5C%22%20style%3D%5C%22float%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%5C%22%3E1.%3C%5C%2Fdiv%3E%3Cdiv%20class%3D%5C%22csl-right-inline%5C%22%20style%3D%5C%22margin%3A%200%20.4em%200%201.5em%3B%5C%22%3EChoi%2C%20K.%2C%20Yeo%2C%20E.%2C%20Chang%2C%20K.%2C%20Watanabe%2C%20S.%20%26amp%3B%20Mortensen%2C%20D.%20Leveraging%20Allophony%20in%20Self-Supervised%20Speech%20Models%20for%20Atypical%20Pronunciation%20Assessment.%20Preprint%20at%20%3Ca%20class%3D%27zp-DOIURL%27%20href%3D%27https%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2502.07029%27%3Ehttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2502.07029%3C%5C%2Fa%3E%20%282025%29.%3C%5C%2Fdiv%3E%5Cn%20%20%3C%5C%2Fdiv%3E%5Cn%3C%5C%2Fdiv%3E%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22Leveraging%20Allophony%20in%20Self-Supervised%20Speech%20Models%20for%20Atypical%20Pronunciation%20Assessment%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Kwanghee%22%2C%22lastName%22%3A%22Choi%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Eunjung%22%2C%22lastName%22%3A%22Yeo%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Kalvin%22%2C%22lastName%22%3A%22Chang%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Shinji%22%2C%22lastName%22%3A%22Watanabe%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22David%22%2C%22lastName%22%3A%22Mortensen%22%7D%5D%2C%22abstractNote%22%3A%22Allophony%20refers%20to%20the%20variation%20in%20the%20phonetic%20realization%20of%20a%20phoneme%20based%20on%20its%20phonetic%20environment.%20Modeling%20allophones%20is%20crucial%20for%20atypical%20pronunciation%20assessment%2C%20which%20involves%20distinguishing%20atypical%20from%20typical%20pronunciations.%20However%2C%20recent%20phoneme%20classifier-based%20approaches%20often%20simplify%20this%20by%20treating%20various%20realizations%20as%20a%20single%20phoneme%2C%20bypassing%20the%20complexity%20of%20modeling%20allophonic%20variation.%20Motivated%20by%20the%20acoustic%20modeling%20capabilities%20of%20frozen%20self-supervised%20speech%20model%20%28S3M%29%20features%2C%20we%20propose%20MixGoP%2C%20a%20novel%20approach%20that%20leverages%20Gaussian%20mixture%20models%20to%20model%20phoneme%20distributions%20with%20multiple%20subclusters.%20Our%20experiments%20show%20that%20MixGoP%20achieves%20state-of-the-art%20performance%20across%20four%20out%20of%20five%20datasets%2C%20including%20dysarthric%20and%20non-native%20speech.%20Our%20analysis%20further%20suggests%20that%20S3M%20features%20capture%20allophonic%20variation%20more%20effectively%20than%20MFCCs%20and%20Mel%20spectrograms%2C%20highlighting%20the%20benefits%20of%20integrating%20MixGoP%20with%20S3M%20features.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222025%22%2C%22DOI%22%3A%2210.48550%5C%2FARXIV.2502.07029%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2502.07029%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222025-03-06T18%3A59%3A43Z%22%7D%7D%2C%7B%22key%22%3A%229TTN32AJ%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Beaglehole%20et%20al.%22%2C%22parsedDate%22%3A%222025%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%3Cdiv%20class%3D%5C%22csl-bib-body%5C%22%20style%3D%5C%22line-height%3A%202%3B%20%5C%22%3E%5Cn%20%20%3Cdiv%20class%3D%5C%22csl-entry%5C%22%20style%3D%5C%22clear%3A%20left%3B%20%5C%22%3E%5Cn%20%20%20%20%3Cdiv%20class%3D%5C%22csl-left-margin%5C%22%20style%3D%5C%22float%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%5C%22%3E1.%3C%5C%2Fdiv%3E%3Cdiv%20class%3D%5C%22csl-right-inline%5C%22%20style%3D%5C%22margin%3A%200%20.4em%200%201.5em%3B%5C%22%3EBeaglehole%2C%20D.%2C%20Radhakrishnan%2C%20A.%2C%20Boix-Adser%26%23xE0%3B%2C%20E.%20%26amp%3B%20Belkin%2C%20M.%20Aggregate%20and%20conquer%3A%20detecting%20and%20steering%20LLM%20concepts%20by%20combining%20nonlinear%20predictors%20over%20multiple%20layers.%20Preprint%20at%20%3Ca%20class%3D%27zp-DOIURL%27%20href%3D%27https%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2502.03708%27%3Ehttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2502.03708%3C%5C%2Fa%3E%20%282025%29.%3C%5C%2Fdiv%3E%5Cn%20%20%3C%5C%2Fdiv%3E%5Cn%3C%5C%2Fdiv%3E%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22Aggregate%20and%20conquer%3A%20detecting%20and%20steering%20LLM%20concepts%20by%20combining%20nonlinear%20predictors%20over%20multiple%20layers%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Daniel%22%2C%22lastName%22%3A%22Beaglehole%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Adityanarayanan%22%2C%22lastName%22%3A%22Radhakrishnan%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Enric%22%2C%22lastName%22%3A%22Boix-Adser%5Cu00e0%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Mikhail%22%2C%22lastName%22%3A%22Belkin%22%7D%5D%2C%22abstractNote%22%3A%22A%20trained%20Large%20Language%20Model%20%28LLM%29%20contains%20much%20of%20human%20knowledge.%20Yet%2C%20it%20is%20difficult%20to%20gauge%20the%20extent%20or%20accuracy%20of%20that%20knowledge%2C%20as%20LLMs%20do%20not%20always%20%60%60know%20what%20they%20know%27%27%20and%20may%20even%20be%20actively%20misleading.%20In%20this%20work%2C%20we%20give%20a%20general%20method%20for%20detecting%20semantic%20concepts%20in%20the%20internal%20activations%20of%20LLMs.%20Furthermore%2C%20we%20show%20that%20our%20methodology%20can%20be%20easily%20adapted%20to%20steer%20LLMs%20toward%20desirable%20outputs.%20Our%20innovations%20are%20the%20following%3A%20%281%29%20we%20use%20a%20nonlinear%20feature%20learning%20method%20to%20identify%20important%20linear%20directions%20for%20predicting%20concepts%20from%20each%20layer%3B%20%282%29%20we%20aggregate%20features%20across%20layers%20to%20build%20powerful%20concept%20detectors%20and%20steering%20mechanisms.%20We%20showcase%20the%20power%20of%20our%20approach%20by%20attaining%20state-of-the-art%20results%20for%20detecting%20hallucinations%2C%20harmfulness%2C%20toxicity%2C%20and%20untruthful%20content%20on%20seven%20benchmarks.%20We%20highlight%20the%20generality%20of%20our%20approach%20by%20steering%20LLMs%20towards%20new%20concepts%20that%2C%20to%20the%20best%20of%20our%20knowledge%2C%20have%20not%20been%20previously%20considered%20in%20the%20literature%2C%20including%3A%20semantic%20disambiguation%2C%20human%20languages%2C%20programming%20languages%2C%20hallucinated%20responses%2C%20science%20subjects%2C%20poetic%5C%2FShakespearean%20English%2C%20and%20even%20multiple%20concepts%20simultaneously.%20Moreover%2C%20our%20method%20can%20steer%20concepts%20with%20numerical%20attributes%20such%20as%20product%20reviews.%20We%20provide%20our%20code%20%28including%20a%20simple%20API%20for%20our%20methods%29%20at%20https%3A%5C%2F%5C%2Fgithub.com%5C%2Fdmbeaglehole%5C%2Fneural_controllers%20.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222025%22%2C%22DOI%22%3A%2210.48550%5C%2FARXIV.2502.03708%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2502.03708%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222025-03-06T18%3A45%3A46Z%22%7D%7D%2C%7B%22key%22%3A%22JU9X74NV%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Richards%20et%20al.%22%2C%22parsedDate%22%3A%222025%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%3Cdiv%20class%3D%5C%22csl-bib-body%5C%22%20style%3D%5C%22line-height%3A%202%3B%20%5C%22%3E%5Cn%20%20%3Cdiv%20class%3D%5C%22csl-entry%5C%22%20style%3D%5C%22clear%3A%20left%3B%20%5C%22%3E%5Cn%20%20%20%20%3Cdiv%20class%3D%5C%22csl-left-margin%5C%22%20style%3D%5C%22float%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%5C%22%3E1.%3C%5C%2Fdiv%3E%3Cdiv%20class%3D%5C%22csl-right-inline%5C%22%20style%3D%5C%22margin%3A%200%20.4em%200%201.5em%3B%5C%22%3ERichards%2C%20C.%2C%20Dima%2C%20A.%2C%20Ferguson%2C%20D.%20%26amp%3B%20Witek%2C%20H.%20Growing%20black-hole%20hair%20in%20nonminimally%20coupled%20biscalar%20gravity.%20Preprint%20at%20%3Ca%20class%3D%27zp-DOIURL%27%20href%3D%27https%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2501.14034%27%3Ehttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2501.14034%3C%5C%2Fa%3E%20%282025%29.%3C%5C%2Fdiv%3E%5Cn%20%20%3C%5C%2Fdiv%3E%5Cn%3C%5C%2Fdiv%3E%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22Growing%20black-hole%20hair%20in%20nonminimally%20coupled%20biscalar%20gravity%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Chloe%22%2C%22lastName%22%3A%22Richards%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Alexandru%22%2C%22lastName%22%3A%22Dima%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Deborah%22%2C%22lastName%22%3A%22Ferguson%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Helvi%22%2C%22lastName%22%3A%22Witek%22%7D%5D%2C%22abstractNote%22%3A%22Black%20holes%20offer%20a%20unique%20laboratory%20for%20fundamental%20physics%20and%20are%20crucial%20for%20probing%20theories%20beyond%20Einstein%27s%20theory%20of%20General%20Relativity.%20In%20this%20paper%2C%20we%20consider%204D%20effective%20field%20theories%20with%20scalar%20fields.%20We%20focus%20on%20axi-dilaton%20gravity%2C%20a%20quadratic%20gravity%20theory%20with%20two%20kinetically%20coupled%20scalar%20fields%2C%20an%20axion%20and%20a%20dilaton.%20To%20evolve%20these%20fields%20around%20black%20holes%2C%20we%20introduce%20Canuda-AxiDil%2C%20the%20first%20open-source%2C%20parameterized%20numerical%20relativity%20code%20for%20quadratic%20and%20bi-scalar%20gravity.%20Using%20this%20code%2C%20we%20perform%20single%20black%20hole%20simulations%20to%20show%20the%20dynamical%20formation%20of%20axion%20and%20dilaton%20hairs.%20Through%20these%20simulations%2C%20we%20measure%20the%20impact%20of%20black-hole%20spin%20and%20curvature%20coupling%20strength%20on%20the%20axion%20and%20dilaton%2C%20and%20show%20that%20a%20kinetic%20coupling%20between%20the%20fields%20increases%20the%20observed%20deviations%20from%20General%20Relativity.%20Furthermore%2C%20we%20simulate%20the%20axion%20and%20dilaton%20fields%20around%20a%20binary%20black%20hole%20coalescence%20demonstrating%20the%20growth%20of%20axion%20hair%20during%20the%20inspiral%20and%20the%20production%20of%20radiative%20modes%20for%20both%20fields.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222025%22%2C%22DOI%22%3A%2210.48550%5C%2FARXIV.2501.14034%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2501.14034%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222025-02-04T22%3A37%3A46Z%22%7D%7D%2C%7B%22key%22%3A%22GXI7LZRB%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Osorio%20et%20al.%22%2C%22parsedDate%22%3A%222024-11-26%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%3Cdiv%20class%3D%5C%22csl-bib-body%5C%22%20style%3D%5C%22line-height%3A%202%3B%20%5C%22%3E%5Cn%20%20%3Cdiv%20class%3D%5C%22csl-entry%5C%22%20style%3D%5C%22clear%3A%20left%3B%20%5C%22%3E%5Cn%20%20%20%20%3Cdiv%20class%3D%5C%22csl-left-margin%5C%22%20style%3D%5C%22float%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%5C%22%3E1.%3C%5C%2Fdiv%3E%3Cdiv%20class%3D%5C%22csl-right-inline%5C%22%20style%3D%5C%22margin%3A%200%20.4em%200%201.5em%3B%5C%22%3EOsorio%2C%20J.%20%3Ci%3Eet%20al.%3C%5C%2Fi%3E%20Keep%20it%20Local%3A%20Comparing%20Domain-Specific%20LLMs%20in%20Native%20and%20Machine%20Translated%20Text%20using%20Parallel%20Corpora%20on%20Political%20Conflict.%20in%20%3Ci%3E2024%202nd%20International%20Conference%20on%20Foundation%20and%20Large%20Language%20Models%20%28FLLM%29%3C%5C%2Fi%3E%20542%26%23x2013%3B552%20%28IEEE%2C%20Dubai%2C%20United%20Arab%20Emirates%2C%202024%29.%20%3Ca%20class%3D%27zp-DOIURL%27%20href%3D%27http%3A%5C%2F%5C%2Fdoi.org%5C%2F10.1109%5C%2FFLLM63129.2024.10852489%27%3Ehttp%3A%5C%2F%5C%2Fdoi.org%5C%2F10.1109%5C%2FFLLM63129.2024.10852489%3C%5C%2Fa%3E.%3C%5C%2Fdiv%3E%5Cn%20%20%3C%5C%2Fdiv%3E%5Cn%3C%5C%2Fdiv%3E%22%2C%22data%22%3A%7B%22itemType%22%3A%22conferencePaper%22%2C%22title%22%3A%22Keep%20it%20Local%3A%20Comparing%20Domain-Specific%20LLMs%20in%20Native%20and%20Machine%20Translated%20Text%20using%20Parallel%20Corpora%20on%20Political%20Conflict%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Javier%22%2C%22lastName%22%3A%22Osorio%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Sultan%22%2C%22lastName%22%3A%22Alsarra%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Amber%22%2C%22lastName%22%3A%22Converse%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Afraa%22%2C%22lastName%22%3A%22Alshammari%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Dagmar%22%2C%22lastName%22%3A%22Heintze%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Latifur%22%2C%22lastName%22%3A%22Khan%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Naif%22%2C%22lastName%22%3A%22Alatrush%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Patrick%20T.%22%2C%22lastName%22%3A%22Brandt%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Vito%22%2C%22lastName%22%3A%22D%5Cu2019Orazio%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Niamat%22%2C%22lastName%22%3A%22Zawad%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Mahrusa%22%2C%22lastName%22%3A%22Billah%22%7D%5D%2C%22abstractNote%22%3A%22%22%2C%22date%22%3A%222024-11-26%22%2C%22proceedingsTitle%22%3A%222024%202nd%20International%20Conference%20on%20Foundation%20and%20Large%20Language%20Models%20%28FLLM%29%22%2C%22conferenceName%22%3A%222024%202nd%20International%20Conference%20on%20Foundation%20and%20Large%20Language%20Models%20%28FLLM%29%22%2C%22language%22%3A%22%22%2C%22DOI%22%3A%2210.1109%5C%2FFLLM63129.2024.10852489%22%2C%22ISBN%22%3A%229798350354799%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Fieeexplore.ieee.org%5C%2Fdocument%5C%2F10852489%5C%2F%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222025-02-04T22%3A29%3A01Z%22%7D%7D%2C%7B%22key%22%3A%22I4F4MX2R%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Avdiunina%20et%20al.%22%2C%22parsedDate%22%3A%222025-01-22%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%3Cdiv%20class%3D%5C%22csl-bib-body%5C%22%20style%3D%5C%22line-height%3A%202%3B%20%5C%22%3E%5Cn%20%20%3Cdiv%20class%3D%5C%22csl-entry%5C%22%20style%3D%5C%22clear%3A%20left%3B%20%5C%22%3E%5Cn%20%20%20%20%3Cdiv%20class%3D%5C%22csl-left-margin%5C%22%20style%3D%5C%22float%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%5C%22%3E1.%3C%5C%2Fdiv%3E%3Cdiv%20class%3D%5C%22csl-right-inline%5C%22%20style%3D%5C%22margin%3A%200%20.4em%200%201.5em%3B%5C%22%3EAvdiunina%2C%20P.%2C%20Jamal%2C%20S.%2C%20Gusev%2C%20F.%20%26amp%3B%20Isayev%2C%20O.%20All%20that%20glitters%20is%20not%20gold%3A%20Importance%20of%20rigorous%20evaluation%20of%20proteochemometric%20models.%20Preprint%20at%20%3Ca%20class%3D%27zp-DOIURL%27%20href%3D%27https%3A%5C%2F%5C%2Fdoi.org%5C%2F10.26434%5C%2Fchemrxiv-2025-vbmgc%27%3Ehttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.26434%5C%2Fchemrxiv-2025-vbmgc%3C%5C%2Fa%3E%20%282025%29.%3C%5C%2Fdiv%3E%5Cn%20%20%3C%5C%2Fdiv%3E%5Cn%3C%5C%2Fdiv%3E%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22All%20that%20glitters%20is%20not%20gold%3A%20Importance%20of%20rigorous%20evaluation%20of%20proteochemometric%20models%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Polina%22%2C%22lastName%22%3A%22Avdiunina%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Shamieraah%22%2C%22lastName%22%3A%22Jamal%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Filipp%22%2C%22lastName%22%3A%22Gusev%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Olexandr%22%2C%22lastName%22%3A%22Isayev%22%7D%5D%2C%22abstractNote%22%3A%22Proteochemometric%20models%20%28PCM%29%20are%20used%20in%20computational%20drug%20discovery%20to%20leverage%20both%20protein%20and%20ligand%20representations%20for%20bioactivity%20prediction.%20While%20machine%20learning%20%28ML%29%20and%20deep%20learning%20%28DL%29%20have%20come%20to%20dominate%20PCMs%2C%20often%20serving%20as%20scoring%20functions%2C%20rigorous%20evaluation%20standards%20have%20not%20always%20been%20consistently%20applied.%20In%20this%20study%2C%20using%20kinase-ligand%20bioactivity%20prediction%20as%20a%20model%20system%2C%20we%20highlight%20the%20critical%20roles%20of%20dataset%20curation%2C%20permutation%20testing%2C%20class%20imbalances%2C%20data%20splitting%20strategies%2C%20and%20embedding%20quality%20in%20determining%20model%20performance.%20Our%20findings%20indicate%20that%20data%20splitting%20and%20class%20imbalances%20are%20the%20most%20critical%20factors%20affecting%20PCM%20performance%2C%20emphasizing%20the%20challenges%20in%20generalizing%20ability%20of%20ML%5C%2FDL-PCMs.%20We%20evaluated%20various%20protein-ligand%20descriptors%20and%20embeddings%2C%20including%20those%20augmented%20with%20multiple%20sequence%20alignment%20%28MSA%29%20information.%20However%2C%20permutation%20testing%20consistently%20demonstrated%20that%20protein%20embeddings%20contributed%20minimally%20to%20PCM%20efficacy.%20This%20study%20advocates%20for%20the%20adoption%20of%20stringent%20evaluation%20standards%20to%20enhance%20the%20generalizability%20of%20models%20to%20out-of-distribution%20data%20and%20improve%20benchmarking%20practices.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222025-01-22%22%2C%22DOI%22%3A%2210.26434%5C%2Fchemrxiv-2025-vbmgc%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Fchemrxiv.org%5C%2Fengage%5C%2Fchemrxiv%5C%2Farticle-details%5C%2F678f32006dde43c908774ef1%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222025-02-04T22%3A17%3A07Z%22%7D%7D%2C%7B%22key%22%3A%22S7DPSCD5%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Pilny%20et%20al.%22%2C%22parsedDate%22%3A%222025-01-22%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%3Cdiv%20class%3D%5C%22csl-bib-body%5C%22%20style%3D%5C%22line-height%3A%202%3B%20%5C%22%3E%5Cn%20%20%3Cdiv%20class%3D%5C%22csl-entry%5C%22%20style%3D%5C%22clear%3A%20left%3B%20%5C%22%3E%5Cn%20%20%20%20%3Cdiv%20class%3D%5C%22csl-left-margin%5C%22%20style%3D%5C%22float%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%5C%22%3E1.%3C%5C%2Fdiv%3E%3Cdiv%20class%3D%5C%22csl-right-inline%5C%22%20style%3D%5C%22margin%3A%200%20.4em%200%201.5em%3B%5C%22%3EPilny%2C%20A.%2C%20Bonito%2C%20J.%20%26amp%3B%20Schecter%2C%20A.%20Coding%20Small%20Group%20Communication%20with%20AI%3A%20RNNs%20and%20Transformers%20with%20Context.%20%3Ci%3ESmall%20Group%20Research%3C%5C%2Fi%3E%2010464964251314196%20%282025%29%20%3Ca%20class%3D%27zp-DOIURL%27%20href%3D%27http%3A%5C%2F%5C%2Fdoi.org%5C%2F10.1177%5C%2F10464964251314197%27%3Ehttp%3A%5C%2F%5C%2Fdoi.org%5C%2F10.1177%5C%2F10464964251314197%3C%5C%2Fa%3E.%3C%5C%2Fdiv%3E%5Cn%20%20%3C%5C%2Fdiv%3E%5Cn%3C%5C%2Fdiv%3E%22%2C%22data%22%3A%7B%22itemType%22%3A%22journalArticle%22%2C%22title%22%3A%22Coding%20Small%20Group%20Communication%20with%20AI%3A%20RNNs%20and%20Transformers%20with%20Context%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Andrew%22%2C%22lastName%22%3A%22Pilny%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Joseph%22%2C%22lastName%22%3A%22Bonito%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Aaron%22%2C%22lastName%22%3A%22Schecter%22%7D%5D%2C%22abstractNote%22%3A%22This%20study%20compares%20the%20performance%20of%20recurrent%20neural%20networks%20%28RNNs%29%20and%20transformer-based%20models%20%28DistilBERT%29%20in%20classifying%20utterances%20as%20dialogue%20acts.%20The%20results%20show%20that%20transformers%20consistently%20outperform%20RNNs%2C%20highlighting%20their%20usefulness%20in%20coding%20small%20group%20interaction.%20Furthermore%2C%20the%20study%20explores%20the%20impact%20of%20incorporating%20context%2C%20in%20the%20form%20of%20preceding%20and%20following%20utterances.%20The%20findings%20reveal%20that%20adding%20context%20leads%20to%20modest%20improvements%20in%20model%20performance.%20Moreover%2C%20in%20some%20cases%2C%20adding%20context%20can%20lead%20to%20a%20slight%20decrease%20in%20performance.%20The%20study%20discusses%20the%20implications%20of%20these%20findings%20for%20small%20group%20researchers%20employing%20AI%20models%20for%20text%20classification%20tasks.%22%2C%22date%22%3A%222025-01-22%22%2C%22language%22%3A%22en%22%2C%22DOI%22%3A%2210.1177%5C%2F10464964251314197%22%2C%22ISSN%22%3A%221046-4964%2C%201552-8278%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Fjournals.sagepub.com%5C%2Fdoi%5C%2F10.1177%5C%2F10464964251314197%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222025-02-04T22%3A15%3A15Z%22%7D%7D%2C%7B%22key%22%3A%22PNIR7KPE%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Deng%20et%20al.%22%2C%22parsedDate%22%3A%222024%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%3Cdiv%20class%3D%5C%22csl-bib-body%5C%22%20style%3D%5C%22line-height%3A%202%3B%20%5C%22%3E%5Cn%20%20%3Cdiv%20class%3D%5C%22csl-entry%5C%22%20style%3D%5C%22clear%3A%20left%3B%20%5C%22%3E%5Cn%20%20%20%20%3Cdiv%20class%3D%5C%22csl-left-margin%5C%22%20style%3D%5C%22float%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%5C%22%3E1.%3C%5C%2Fdiv%3E%3Cdiv%20class%3D%5C%22csl-right-inline%5C%22%20style%3D%5C%22margin%3A%200%20.4em%200%201.5em%3B%5C%22%3EDeng%2C%20J.%20%3Ci%3Eet%20al.%3C%5C%2Fi%3E%20%24%5C%5Ctexttt%7Bdattri%7D%24%3A%20A%20Library%20for%20Efficient%20Data%20Attribution.%20Preprint%20at%20%3Ca%20class%3D%27zp-DOIURL%27%20href%3D%27https%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2410.04555%27%3Ehttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2410.04555%3C%5C%2Fa%3E%20%282024%29.%3C%5C%2Fdiv%3E%5Cn%20%20%3C%5C%2Fdiv%3E%5Cn%3C%5C%2Fdiv%3E%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22%24%5C%5Ctexttt%7Bdattri%7D%24%3A%20A%20Library%20for%20Efficient%20Data%20Attribution%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Junwei%22%2C%22lastName%22%3A%22Deng%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Ting-Wei%22%2C%22lastName%22%3A%22Li%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Shiyuan%22%2C%22lastName%22%3A%22Zhang%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Shixuan%22%2C%22lastName%22%3A%22Liu%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Yijun%22%2C%22lastName%22%3A%22Pan%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Hao%22%2C%22lastName%22%3A%22Huang%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Xinhe%22%2C%22lastName%22%3A%22Wang%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Pingbang%22%2C%22lastName%22%3A%22Hu%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Xingjian%22%2C%22lastName%22%3A%22Zhang%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Jiaqi%20W.%22%2C%22lastName%22%3A%22Ma%22%7D%5D%2C%22abstractNote%22%3A%22Data%20attribution%20methods%20aim%20to%20quantify%20the%20influence%20of%20individual%20training%20samples%20on%20the%20prediction%20of%20artificial%20intelligence%20%28AI%29%20models.%20As%20training%20data%20plays%20an%20increasingly%20crucial%20role%20in%20the%20modern%20development%20of%20large-scale%20AI%20models%2C%20data%20attribution%20has%20found%20broad%20applications%20in%20improving%20AI%20performance%20and%20safety.%20However%2C%20despite%20a%20surge%20of%20new%20data%20attribution%20methods%20being%20developed%20recently%2C%20there%20lacks%20a%20comprehensive%20library%20that%20facilitates%20the%20development%2C%20benchmarking%2C%20and%20deployment%20of%20different%20data%20attribution%20methods.%20In%20this%20work%2C%20we%20introduce%20%24%5C%5Ctexttt%7Bdattri%7D%24%2C%20an%20open-source%20data%20attribution%20library%20that%20addresses%20the%20above%20needs.%20Specifically%2C%20%24%5C%5Ctexttt%7Bdattri%7D%24%20highlights%20three%20novel%20design%20features.%20Firstly%2C%20%24%5C%5Ctexttt%7Bdattri%7D%24%20proposes%20a%20unified%20and%20easy-to-use%20API%2C%20allowing%20users%20to%20integrate%20different%20data%20attribution%20methods%20into%20their%20PyTorch-based%20machine%20learning%20pipeline%20with%20a%20few%20lines%20of%20code%20changed.%20Secondly%2C%20%24%5C%5Ctexttt%7Bdattri%7D%24%20modularizes%20low-level%20utility%20functions%20that%20are%20commonly%20used%20in%20data%20attribution%20methods%2C%20such%20as%20Hessian-vector%20product%2C%20inverse-Hessian-vector%20product%20or%20random%20projection%2C%20making%20it%20easier%20for%20researchers%20to%20develop%20new%20data%20attribution%20methods.%20Thirdly%2C%20%24%5C%5Ctexttt%7Bdattri%7D%24%20provides%20a%20comprehensive%20benchmark%20framework%20with%20pre-trained%20models%20and%20ground%20truth%20annotations%20for%20a%20variety%20of%20benchmark%20settings%2C%20including%20generative%20AI%20settings.%20We%20have%20implemented%20a%20variety%20of%20state-of-the-art%20efficient%20data%20attribution%20methods%20that%20can%20be%20applied%20to%20large-scale%20neural%20network%20models%2C%20and%20will%20continuously%20update%20the%20library%20in%20the%20future.%20Using%20the%20developed%20%24%5C%5Ctexttt%7Bdattri%7D%24%20library%2C%20we%20are%20able%20to%20perform%20a%20comprehensive%20and%20fair%20benchmark%20analysis%20across%20a%20wide%20range%20of%20data%20attribution%20methods.%20The%20source%20code%20of%20%24%5C%5Ctexttt%7Bdattri%7D%24%20is%20available%20at%20https%3A%5C%2F%5C%2Fgithub.com%5C%2FTRAIS-Lab%5C%2Fdattri.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222024%22%2C%22DOI%22%3A%2210.48550%5C%2FARXIV.2410.04555%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2410.04555%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222025-02-04T22%3A10%3A23Z%22%7D%7D%2C%7B%22key%22%3A%22G3GYIZP8%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Chen%20et%20al.%22%2C%22parsedDate%22%3A%222024-12-02%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%3Cdiv%20class%3D%5C%22csl-bib-body%5C%22%20style%3D%5C%22line-height%3A%202%3B%20%5C%22%3E%5Cn%20%20%3Cdiv%20class%3D%5C%22csl-entry%5C%22%20style%3D%5C%22clear%3A%20left%3B%20%5C%22%3E%5Cn%20%20%20%20%3Cdiv%20class%3D%5C%22csl-left-margin%5C%22%20style%3D%5C%22float%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%5C%22%3E1.%3C%5C%2Fdiv%3E%3Cdiv%20class%3D%5C%22csl-right-inline%5C%22%20style%3D%5C%22margin%3A%200%20.4em%200%201.5em%3B%5C%22%3EChen%2C%20W.%2C%20Yan%2C%20B.%2C%20Chen%2C%20C.-C.%20%26amp%3B%20Watanabe%2C%20S.%20Floras%2050%3A%20A%20Massively%20Multilingual%20Multitask%20Benchmark%20for%20Long-Form%20Conversational%20Speech.%20in%20%3Ci%3E2024%20IEEE%20Spoken%20Language%20Technology%20Workshop%20%28SLT%29%3C%5C%2Fi%3E%20891%26%23x2013%3B898%20%28IEEE%2C%20Macao%2C%202024%29.%20%3Ca%20class%3D%27zp-DOIURL%27%20href%3D%27http%3A%5C%2F%5C%2Fdoi.org%5C%2F10.1109%5C%2FSLT61566.2024.10832167%27%3Ehttp%3A%5C%2F%5C%2Fdoi.org%5C%2F10.1109%5C%2FSLT61566.2024.10832167%3C%5C%2Fa%3E.%3C%5C%2Fdiv%3E%5Cn%20%20%3C%5C%2Fdiv%3E%5Cn%3C%5C%2Fdiv%3E%22%2C%22data%22%3A%7B%22itemType%22%3A%22conferencePaper%22%2C%22title%22%3A%22Floras%2050%3A%20A%20Massively%20Multilingual%20Multitask%20Benchmark%20for%20Long-Form%20Conversational%20Speech%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22William%22%2C%22lastName%22%3A%22Chen%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Brian%22%2C%22lastName%22%3A%22Yan%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Chih-Chen%22%2C%22lastName%22%3A%22Chen%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Shinji%22%2C%22lastName%22%3A%22Watanabe%22%7D%5D%2C%22abstractNote%22%3A%22%22%2C%22date%22%3A%222024-12-2%22%2C%22proceedingsTitle%22%3A%222024%20IEEE%20Spoken%20Language%20Technology%20Workshop%20%28SLT%29%22%2C%22conferenceName%22%3A%222024%20IEEE%20Spoken%20Language%20Technology%20Workshop%20%28SLT%29%22%2C%22language%22%3A%22%22%2C%22DOI%22%3A%2210.1109%5C%2FSLT61566.2024.10832167%22%2C%22ISBN%22%3A%229798350392258%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Fieeexplore.ieee.org%5C%2Fdocument%5C%2F10832167%5C%2F%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222025-02-04T21%3A36%3A11Z%22%7D%7D%2C%7B%22key%22%3A%22LWNI38D2%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Nakamura%20et%20al.%22%2C%22parsedDate%22%3A%222025%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%3Cdiv%20class%3D%5C%22csl-bib-body%5C%22%20style%3D%5C%22line-height%3A%202%3B%20%5C%22%3E%5Cn%20%20%3Cdiv%20class%3D%5C%22csl-entry%5C%22%20style%3D%5C%22clear%3A%20left%3B%20%5C%22%3E%5Cn%20%20%20%20%3Cdiv%20class%3D%5C%22csl-left-margin%5C%22%20style%3D%5C%22float%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%5C%22%3E1.%3C%5C%2Fdiv%3E%3Cdiv%20class%3D%5C%22csl-right-inline%5C%22%20style%3D%5C%22margin%3A%200%20.4em%200%201.5em%3B%5C%22%3ENakamura%2C%20T.%20%3Ci%3Eet%20al.%3C%5C%2Fi%3E%20Discrete%20Speech%20Unit%20Extraction%20via%20Independent%20Component%20Analysis.%20Preprint%20at%20%3Ca%20class%3D%27zp-DOIURL%27%20href%3D%27https%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2501.06562%27%3Ehttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2501.06562%3C%5C%2Fa%3E%20%282025%29.%3C%5C%2Fdiv%3E%5Cn%20%20%3C%5C%2Fdiv%3E%5Cn%3C%5C%2Fdiv%3E%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22Discrete%20Speech%20Unit%20Extraction%20via%20Independent%20Component%20Analysis%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Tomohiko%22%2C%22lastName%22%3A%22Nakamura%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Kwanghee%22%2C%22lastName%22%3A%22Choi%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Keigo%22%2C%22lastName%22%3A%22Hojo%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Yoshiaki%22%2C%22lastName%22%3A%22Bando%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Satoru%22%2C%22lastName%22%3A%22Fukayama%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Shinji%22%2C%22lastName%22%3A%22Watanabe%22%7D%5D%2C%22abstractNote%22%3A%22Self-supervised%20speech%20models%20%28S3Ms%29%20have%20become%20a%20common%20tool%20for%20the%20speech%20processing%20community%2C%20leveraging%20representations%20for%20downstream%20tasks.%20Clustering%20S3M%20representations%20yields%20discrete%20speech%20units%20%28DSUs%29%2C%20which%20serve%20as%20compact%20representations%20for%20speech%20signals.%20DSUs%20are%20typically%20obtained%20by%20k-means%20clustering.%20Using%20DSUs%20often%20leads%20to%20strong%20performance%20in%20various%20tasks%2C%20including%20automatic%20speech%20recognition%20%28ASR%29.%20However%2C%20even%20with%20the%20high%20dimensionality%20and%20redundancy%20of%20S3M%20representations%2C%20preprocessing%20S3M%20representations%20for%20better%20clustering%20remains%20unexplored%2C%20even%20though%20it%20can%20affect%20the%20quality%20of%20DSUs.%20In%20this%20paper%2C%20we%20investigate%20the%20potential%20of%20linear%20preprocessing%20methods%20for%20extracting%20DSUs.%20We%20evaluate%20standardization%2C%20principal%20component%20analysis%2C%20whitening%2C%20and%20independent%20component%20analysis%20%28ICA%29%20on%20DSU-based%20ASR%20benchmarks%20and%20demonstrate%20their%20effectiveness%20as%20preprocessing%20for%20k-means.%20We%20also%20conduct%20extensive%20analyses%20of%20their%20behavior%2C%20such%20as%20orthogonality%20or%20interpretability%20of%20individual%20components%20of%20ICA.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222025%22%2C%22DOI%22%3A%2210.48550%5C%2FARXIV.2501.06562%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2501.06562%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222025-02-04T16%3A19%3A02Z%22%7D%7D%2C%7B%22key%22%3A%22CGS26BTG%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Khot%20et%20al.%22%2C%22parsedDate%22%3A%222025%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%3Cdiv%20class%3D%5C%22csl-bib-body%5C%22%20style%3D%5C%22line-height%3A%202%3B%20%5C%22%3E%5Cn%20%20%3Cdiv%20class%3D%5C%22csl-entry%5C%22%20style%3D%5C%22clear%3A%20left%3B%20%5C%22%3E%5Cn%20%20%20%20%3Cdiv%20class%3D%5C%22csl-left-margin%5C%22%20style%3D%5C%22float%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%5C%22%3E1.%3C%5C%2Fdiv%3E%3Cdiv%20class%3D%5C%22csl-right-inline%5C%22%20style%3D%5C%22margin%3A%200%20.4em%200%201.5em%3B%5C%22%3EKhot%2C%20A.%2C%20Wang%2C%20X.%2C%20Roy%2C%20A.%2C%20Kindratenko%2C%20V.%20%26amp%3B%20Neubauer%2C%20M.%20S.%20Evidential%20Deep%20Learning%20for%20Uncertainty%20Quantification%20and%20Out-of-Distribution%20Detection%20in%20Jet%20Identification%20using%20Deep%20Neural%20Networks.%20Preprint%20at%20%3Ca%20class%3D%27zp-DOIURL%27%20href%3D%27https%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2501.05656%27%3Ehttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2501.05656%3C%5C%2Fa%3E%20%282025%29.%3C%5C%2Fdiv%3E%5Cn%20%20%3C%5C%2Fdiv%3E%5Cn%3C%5C%2Fdiv%3E%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22Evidential%20Deep%20Learning%20for%20Uncertainty%20Quantification%20and%20Out-of-Distribution%20Detection%20in%20Jet%20Identification%20using%20Deep%20Neural%20Networks%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Ayush%22%2C%22lastName%22%3A%22Khot%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Xiwei%22%2C%22lastName%22%3A%22Wang%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Avik%22%2C%22lastName%22%3A%22Roy%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Volodymyr%22%2C%22lastName%22%3A%22Kindratenko%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Mark%20S.%22%2C%22lastName%22%3A%22Neubauer%22%7D%5D%2C%22abstractNote%22%3A%22Current%20methods%20commonly%20used%20for%20uncertainty%20quantification%20%28UQ%29%20in%20deep%20learning%20%28DL%29%20models%20utilize%20Bayesian%20methods%20which%20are%20computationally%20expensive%20and%20time-consuming.%20In%20this%20paper%2C%20we%20provide%20a%20detailed%20study%20of%20UQ%20based%20on%20evidential%20deep%20learning%20%28EDL%29%20for%20deep%20neural%20network%20models%20designed%20to%20identify%20jets%20in%20high%20energy%20proton-proton%20collisions%20at%20the%20Large%20Hadron%20Collider%20and%20explore%20its%20utility%20in%20anomaly%20detection.%20EDL%20is%20a%20DL%20approach%20that%20treats%20learning%20as%20an%20evidence%20acquisition%20process%20designed%20to%20provide%20confidence%20%28or%20epistemic%20uncertainty%29%20about%20test%20data.%20Using%20publicly%20available%20datasets%20for%20jet%20classification%20benchmarking%2C%20we%20explore%20hyperparameter%20optimizations%20for%20EDL%20applied%20to%20the%20challenge%20of%20UQ%20for%20jet%20identification.%20We%20also%20investigate%20how%20the%20uncertainty%20is%20distributed%20for%20each%20jet%20class%2C%20how%20this%20method%20can%20be%20implemented%20for%20the%20detection%20of%20anomalies%2C%20how%20the%20uncertainty%20compares%20with%20Bayesian%20ensemble%20methods%2C%20and%20how%20the%20uncertainty%20maps%20onto%20latent%20spaces%20for%20the%20models.%20Our%20studies%20uncover%20some%20pitfalls%20of%20EDL%20applied%20to%20anomaly%20detection%20and%20a%20more%20effective%20way%20to%20quantify%20uncertainty%20from%20EDL%20as%20compared%20with%20the%20foundational%20EDL%20setup.%20These%20studies%20illustrate%20a%20methodological%20approach%20to%20interpreting%20EDL%20in%20jet%20classification%20models%2C%20providing%20new%20insights%20on%20how%20EDL%20quantifies%20uncertainty%20and%20detects%20out-of-distribution%20data%20which%20may%20lead%20to%20improved%20EDL%20methods%20for%20DL%20models%20applied%20to%20classification%20tasks.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222025%22%2C%22DOI%22%3A%2210.48550%5C%2FARXIV.2501.05656%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2501.05656%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222025-02-04T16%3A18%3A46Z%22%7D%7D%2C%7B%22key%22%3A%22MY5DUZTJ%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Andrews%20et%20al.%22%2C%22parsedDate%22%3A%222025-01-23%22%2C%22numChildren%22%3A1%7D%2C%22bib%22%3A%22%3Cdiv%20class%3D%5C%22csl-bib-body%5C%22%20style%3D%5C%22line-height%3A%202%3B%20%5C%22%3E%5Cn%20%20%3Cdiv%20class%3D%5C%22csl-entry%5C%22%20style%3D%5C%22clear%3A%20left%3B%20%5C%22%3E%5Cn%20%20%20%20%3Cdiv%20class%3D%5C%22csl-left-margin%5C%22%20style%3D%5C%22float%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%5C%22%3E1.%3C%5C%2Fdiv%3E%3Cdiv%20class%3D%5C%22csl-right-inline%5C%22%20style%3D%5C%22margin%3A%200%20.4em%200%201.5em%3B%5C%22%3EAndrews%2C%20J.%2C%20Weirich%2C%20K.%20%26amp%3B%20Schiller%2C%20U.%20D.%20Molecular-Scale%20Simulation%20of%20Wetting%20of%20Actin%20Filaments%20by%20Protein%20Droplets.%20%3Ci%3EJ.%20Phys.%20Chem.%20B%3C%5C%2Fi%3E%20%3Cb%3E129%3C%5C%2Fb%3E%2C%201109%26%23x2013%3B1121%20%282025%29.%3C%5C%2Fdiv%3E%5Cn%20%20%3C%5C%2Fdiv%3E%5Cn%3C%5C%2Fdiv%3E%22%2C%22data%22%3A%7B%22itemType%22%3A%22journalArticle%22%2C%22title%22%3A%22Molecular-Scale%20Simulation%20of%20Wetting%20of%20Actin%20Filaments%20by%20Protein%20Droplets%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22James%22%2C%22lastName%22%3A%22Andrews%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Kimberly%22%2C%22lastName%22%3A%22Weirich%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Ulf%20D.%22%2C%22lastName%22%3A%22Schiller%22%7D%5D%2C%22abstractNote%22%3A%22%22%2C%22date%22%3A%222025-01-23%22%2C%22language%22%3A%22en%22%2C%22DOI%22%3A%2210.1021%5C%2Facs.jpcb.4c07282%22%2C%22ISSN%22%3A%221520-6106%2C%201520-5207%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Fpubs.acs.org%5C%2Fdoi%5C%2F10.1021%5C%2Facs.jpcb.4c07282%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222025-02-04T16%3A18%3A26Z%22%7D%7D%2C%7B%22key%22%3A%22J5M5FIB7%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Wang%20et%20al.%22%2C%22parsedDate%22%3A%222025-01-10%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%3Cdiv%20class%3D%5C%22csl-bib-body%5C%22%20style%3D%5C%22line-height%3A%202%3B%20%5C%22%3E%5Cn%20%20%3Cdiv%20class%3D%5C%22csl-entry%5C%22%20style%3D%5C%22clear%3A%20left%3B%20%5C%22%3E%5Cn%20%20%20%20%3Cdiv%20class%3D%5C%22csl-left-margin%5C%22%20style%3D%5C%22float%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%5C%22%3E1.%3C%5C%2Fdiv%3E%3Cdiv%20class%3D%5C%22csl-right-inline%5C%22%20style%3D%5C%22margin%3A%200%20.4em%200%201.5em%3B%5C%22%3EWang%2C%20S.%20%3Ci%3Eet%20al.%3C%5C%2Fi%3E%20Deep%20CNN-based%20semi-supervised%20learning%20approach%20for%20identifying%20and%20segmenting%20corrosion%20in%20hydraulic%20steel%20and%20water%20resources%20infrastructure.%20%3Ci%3EStructural%20Health%20Monitoring%3C%5C%2Fi%3E%2014759217241305040%20%282025%29%20%3Ca%20class%3D%27zp-DOIURL%27%20href%3D%27http%3A%5C%2F%5C%2Fdoi.org%5C%2F10.1177%5C%2F14759217241305039%27%3Ehttp%3A%5C%2F%5C%2Fdoi.org%5C%2F10.1177%5C%2F14759217241305039%3C%5C%2Fa%3E.%3C%5C%2Fdiv%3E%5Cn%20%20%3C%5C%2Fdiv%3E%5Cn%3C%5C%2Fdiv%3E%22%2C%22data%22%3A%7B%22itemType%22%3A%22journalArticle%22%2C%22title%22%3A%22Deep%20CNN-based%20semi-supervised%20learning%20approach%20for%20identifying%20and%20segmenting%20corrosion%20in%20hydraulic%20steel%20and%20water%20resources%20infrastructure%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Shengyi%22%2C%22lastName%22%3A%22Wang%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Hai%22%2C%22lastName%22%3A%22Nguyen%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Rebekah%22%2C%22lastName%22%3A%22Wilson%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Brian%22%2C%22lastName%22%3A%22Eick%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Nora%22%2C%22lastName%22%3A%22El-Gohary%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Carolyn%22%2C%22lastName%22%3A%22Ortiz%22%7D%5D%2C%22abstractNote%22%3A%22The%20United%20States%20faces%20significant%20challenges%20due%20to%20corrosion%2C%20with%20its%20impact%20on%20military%20and%20civilian%20infrastructure%20incurring%20over%20%2420%20billion%20in%20annual%20maintenance%20costs.%20The%20damage%20due%20to%20corrosion%20is%20profound%2C%20threatening%20structural%20safety%2C%20reducing%20esthetic%20value%2C%20and%20leading%20to%20costly%20repairs.%20To%20mitigate%20these%20effects%2C%20the%20Unified%20Facilities%20Criteria%20and%20Unified%20Facilities%20Guidance%20Specifications%20advise%20the%20use%20of%20protective%20coatings%20on%20metal%20surfaces.%20Early%20corrosion%20detection%20is%20crucial%20for%20maintaining%20structural%20integrity%20and%20minimizing%20maintenance%20costs.%20Recent%20breakthroughs%20in%20artificial%20intelligence%20and%20deep%20learning%2C%20including%20accurate%20corrosion%20classification%2C%20have%20significantly%20revolutionized%20the%20detection%20and%20management%20of%20corrosion.%20Despite%20these%20advancements%2C%20automatic%20corrosion%20segmentation%20in%20civil%20infrastructure%20remains%20challenging%20due%20to%20the%20scarcity%20of%20images%20and%20the%20labor-intensive%20annotation%20process.%20Moreover%2C%20existing%20segmentation%20methods%20are%20unable%20to%20manage%20the%20complexities%20that%20come%20with%20high-resolution%20corrosion%20images.%20This%20paper%20proposes%20a%20novel%2C%20semi-supervised%2C%20convolutional%20neural%20network-based%20image%20segmentation%20method%20for%20the%20automatic%20identification%20and%20segmentation%20of%20corrosion%20on%20coated%20steel%20surfaces%2C%20using%20both%20unlabeled%20and%20labeled%20corrosion%20images%20and%20leveraging%20the%20mean%20teacher%20model.%20The%20proposed%20novel%20method%20involves%20three%20steps%3A%20%281%29%20utilizing%20high-resolution%20digital%20microscopy%20to%20capture%20detailed%20images%20and%20dividing%20them%20into%20manageable%20patches%3B%20%282%29%20applying%20a%20semi-supervised%20learning%20approach%2C%20leveraging%20unlabeled%20corrosion%20images%20for%20enhanced%20segmentation%20precision%3B%20and%20%283%29%20employing%20a%20smoothing%20module%20to%20improve%20the%20continuity%20of%20information.%20The%20proposed%20corrosion%20detection%20method%20has%20demonstrated%20promising%20performance%20with%20only%2067%25%20labeled%20data%2C%20achieving%20mean%20precision%2C%20recall%2C%20F-1%20measure%2C%20and%20intersection%20over%20union%20of%2090.0%25%2C%2096.2%25%2C%2092.7%25%2C%20and%2087.1%25%2C%20respectively.%20Even%20with%20just%2033%25%20labeled%20data%2C%20the%20method%20maintains%20strong%20performance%20when%20compared%20to%20fully%20supervised%20deep%20learning%20models.%20This%20demonstrates%20a%20substantial%20data%20resource%20saving%20while%20ensuring%20accurate%20and%20reliable%20corrosion%20detection%2C%20which%20is%20crucial%20for%20infrastructure%20health%20monitoring.%20The%20successful%20validation%20of%20this%20approach%20provides%20a%20method%20that%20dramatically%20reduces%20the%20amount%20of%20visual%20data%20required%20to%20generate%20a%20reliable%20model.%22%2C%22date%22%3A%222025-01-10%22%2C%22language%22%3A%22en%22%2C%22DOI%22%3A%2210.1177%5C%2F14759217241305039%22%2C%22ISSN%22%3A%221475-9217%2C%201741-3168%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Fjournals.sagepub.com%5C%2Fdoi%5C%2F10.1177%5C%2F14759217241305039%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222025-02-04T16%3A18%3A11Z%22%7D%7D%2C%7B%22key%22%3A%22SHNFFNVK%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Feng%20et%20al.%22%2C%22parsedDate%22%3A%222025-01-05%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%3Cdiv%20class%3D%5C%22csl-bib-body%5C%22%20style%3D%5C%22line-height%3A%202%3B%20%5C%22%3E%5Cn%20%20%3Cdiv%20class%3D%5C%22csl-entry%5C%22%20style%3D%5C%22clear%3A%20left%3B%20%5C%22%3E%5Cn%20%20%20%20%3Cdiv%20class%3D%5C%22csl-left-margin%5C%22%20style%3D%5C%22float%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%5C%22%3E1.%3C%5C%2Fdiv%3E%3Cdiv%20class%3D%5C%22csl-right-inline%5C%22%20style%3D%5C%22margin%3A%200%20.4em%200%201.5em%3B%5C%22%3EFeng%2C%20J.%20T.%2C%20Satheesan%2C%20S.%20P.%2C%20Kong%2C%20S.%2C%20Donders%2C%20T.%20H.%20%26amp%3B%20Punyasena%2C%20S.%20W.%20Addressing%20the%20open%20world%3A%20detecting%20and%20segmenting%20pollen%20on%20palynological%20slides%20with%20deep%20learning.%20Preprint%20at%20%3Ca%20class%3D%27zp-DOIURL%27%20href%3D%27https%3A%5C%2F%5C%2Fdoi.org%5C%2F10.1101%5C%2F2025.01.05.631390%27%3Ehttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.1101%5C%2F2025.01.05.631390%3C%5C%2Fa%3E%20%282025%29.%3C%5C%2Fdiv%3E%5Cn%20%20%3C%5C%2Fdiv%3E%5Cn%3C%5C%2Fdiv%3E%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22Addressing%20the%20open%20world%3A%20detecting%20and%20segmenting%20pollen%20on%20palynological%20slides%20with%20deep%20learning%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Jennifer%20T.%22%2C%22lastName%22%3A%22Feng%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Sandeep%20Puthanveetil%22%2C%22lastName%22%3A%22Satheesan%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Shu%22%2C%22lastName%22%3A%22Kong%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Timme%20H.%22%2C%22lastName%22%3A%22Donders%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Surangi%20W.%22%2C%22lastName%22%3A%22Punyasena%22%7D%5D%2C%22abstractNote%22%3A%22In%20the%20open%20world%2C%20categorical%20classes%20are%20imbalanced%2C%20test%20classes%20are%20not%20known%20a%20priori%2C%20and%20test%20data%20are%20captured%20across%20different%20domains.%20Paleontological%20data%20can%20be%20described%20as%20open-world%2C%20as%20specimens%20may%20include%20new%2C%20unknown%20taxa%2C%20and%20the%20data%20collected%2C%20such%20as%20measurements%20or%20images%2C%20may%20not%20be%20standardized%20across%20different%20studies.%20Fossil%20pollen%20analysis%20is%20one%20example%20of%20an%20open-world%20problem%20in%20paleontology.%20Pollen%20samples%20capture%20large%20numbers%20of%20specimens%2C%20including%20not%20only%20common%20types%20but%20also%20rare%20and%20even%20novel%20taxa.%20Pollen%20is%20diverse%20morphologically%20and%20features%20can%20be%20altered%20during%20fossilization.%20Additionally%2C%20there%20is%20little%20standardization%20in%20the%20methods%20used%20to%20capture%20and%20catalog%20pollen%20images%20and%20most%20collections%20are%20mounted%20on%20microscope%20slides.%20Therefore%2C%20generalized%20workflows%20for%20automated%20pollen%20analysis%20require%20techniques%20that%20are%20robust%20to%20these%20differences%20and%20can%20work%20with%20microscope%20images.%20We%20focus%20on%20a%20critical%20first%20step%2C%20the%20detection%20of%20pollen%20specimens%20on%20a%20palynological%20slide%20and%20review%20how%20existing%20methods%20can%20be%20employed%20to%20build%20robust%20and%20generalizable%20analysis%20pipelines.%20First%2C%20we%20demonstrate%20how%20a%20mixture-of-experts%20approach%20--%20the%20fusion%20of%20a%20general%20pollen%20detector%20with%20an%20expert%20model%20trained%20on%20minority%20classes%20--%20can%20be%20used%20to%20address%20taxonomic%20biases%20in%20detections%2C%20particularly%20the%20missed%20detections%20of%20rarer%20pollen%20types.%20Second%2C%20we%20demonstrate%20the%20efficiency%20of%20domain%20fine-tuning%20in%20addressing%20domain%20gaps%20--%20differences%20in%20image%20magnification%20and%20resolution%20across%20microscopes%2C%20and%20of%20taxa%20across%20different%20sample%20sources.%20Third%2C%20we%20demonstrate%20the%20importance%20of%20continual%20learning%20workflows%2C%20which%20integrate%20expert%20feedback%2C%20in%20training%20detection%20models%20from%20incomplete%20data.%20Finally%2C%20we%20demonstrate%20how%20cutting-edge%20segmentation%20models%20can%20be%20used%20to%20refine%20and%20clean%20detections%20for%20downstream%20deep%20learning%20classification%20models.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222025-01-05%22%2C%22DOI%22%3A%2210.1101%5C%2F2025.01.05.631390%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22http%3A%5C%2F%5C%2Fbiorxiv.org%5C%2Flookup%5C%2Fdoi%5C%2F10.1101%5C%2F2025.01.05.631390%22%2C%22language%22%3A%22en%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222025-02-04T16%3A17%3A45Z%22%7D%7D%2C%7B%22key%22%3A%22FMC7GJBX%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Vatansever%20and%20Levin%22%2C%22parsedDate%22%3A%222025%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%3Cdiv%20class%3D%5C%22csl-bib-body%5C%22%20style%3D%5C%22line-height%3A%202%3B%20%5C%22%3E%5Cn%20%20%3Cdiv%20class%3D%5C%22csl-entry%5C%22%20style%3D%5C%22clear%3A%20left%3B%20%5C%22%3E%5Cn%20%20%20%20%3Cdiv%20class%3D%5C%22csl-left-margin%5C%22%20style%3D%5C%22float%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%5C%22%3E1.%3C%5C%2Fdiv%3E%3Cdiv%20class%3D%5C%22csl-right-inline%5C%22%20style%3D%5C%22margin%3A%200%20.4em%200%201.5em%3B%5C%22%3EVatansever%2C%20D.%20%26amp%3B%20Levin%2C%20D.%20Collisionless%20Plasma%20Plume%20Expansion%20Under%20External%20Magnetic%20Fields.%20%282025%29.%3C%5C%2Fdiv%3E%5Cn%20%20%3C%5C%2Fdiv%3E%5Cn%3C%5C%2Fdiv%3E%22%2C%22data%22%3A%7B%22itemType%22%3A%22presentation%22%2C%22title%22%3A%22Collisionless%20Plasma%20Plume%20Expansion%20Under%20External%20Magnetic%20Fields%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22presenter%22%2C%22firstName%22%3A%22Davut%22%2C%22lastName%22%3A%22Vatansever%22%7D%2C%7B%22creatorType%22%3A%22presenter%22%2C%22firstName%22%3A%22Deborah%22%2C%22lastName%22%3A%22Levin%22%7D%5D%2C%22abstractNote%22%3A%22%22%2C%22date%22%3A%222025%2C%2001%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Farc.aiaa.org%5C%2Fdoi%5C%2F10.2514%5C%2F6.2025-2491%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222025-02-04T16%3A17%3A19Z%22%7D%7D%5D%7D
1.
Rapp, J. et al. Design of Tough 3D Printable Elastomers with Human-in-the-Loop Reinforcement Learning. Preprint at https://doi.org/10.26434/chemrxiv-2025-w1563 (2025).
1.
Yang, Y., Taherian, H., Kalkhorani, V. A. & Wang, D. Elevating Robust Multi-Talker ASR by Decoupling Speaker Separation and Speech Recognition. Preprint at https://doi.org/10.48550/ARXIV.2503.17886 (2025).
1.
Yan, J. & Snir, M. Contemplating a Lightweight Communication Interface for Asynchronous Many-Task Systems. Preprint at https://doi.org/10.48550/ARXIV.2503.15400 (2025).
1.
Huang, H.-K., Park, S., Villa, U., Wang, L. V. & Anastasio, M. A. Gradient-free joint reconstruction of initial pressure distribution and wave speeds in transcranial photoacoustic computed tomography. in Photons Plus Ultrasound: Imaging and Sensing 2025 vol. 13319 97–103 (SPIE, 2025).
1.
Pant, S. et al. Dissecting Large-Scale Structural Transitions in Membrane Transporters Using Advanced Simulation Technologies. J. Phys. Chem. B acs.jpcb.5c00104 (2025) http://doi.org/10.1021/acs.jpcb.5c00104.
1.
Hu, Y. et al. Empirical Privacy Variance. Preprint at https://doi.org/10.48550/ARXIV.2503.12314 (2025).
1.
Chung, A. K.-W., Lam, K. K.-H. & Yunes, N. Quasinormal mode frequencies and gravitational perturbations of spinning black holes in modified gravity through METRICS: The dynamical Chern-Simons gravity case. Preprint at https://doi.org/10.48550/ARXIV.2503.11759 (2025).
1.
Yan, J., Kaiser, H. & Snir, M. Understanding the Communication Needs of Asynchronous Many-Task Systems -- A Case Study of HPX+LCI. Preprint at https://doi.org/10.48550/ARXIV.2503.12774 (2025).
1.
Merzky, A. et al. Scalable Runtime Architecture for Data-driven, Hybrid HPC and ML Workflow Applications. Preprint at https://doi.org/10.48550/ARXIV.2503.13343 (2025).
1.
Cui, S. et al. Characterizing GPU Resilience and Impact on AI/HPC Systems. Preprint at https://doi.org/10.48550/ARXIV.2503.11901 (2025).
1.
You, Z. & Guo, Y. PlainQAFact: Automatic Factuality Evaluation Metric for Biomedical Plain Language Summaries Generation. Preprint at https://doi.org/10.48550/ARXIV.2503.08890 (2025).
1.
Balaji, P. et al. Quantum Circuits for SU(3) Lattice Gauge Theory. Preprint at https://doi.org/10.48550/ARXIV.2503.08866 (2025).
1.
Sasidharan, A., Xian-He, Lofstead, J. & Klasky, S. Performance Models for a Two-tiered Storage System. Preprint at https://doi.org/10.48550/ARXIV.2503.08966 (2025).
1.
Prather, B. S. KHARMA: Flexible, Portable Performance for GRMHD. Preprint at https://doi.org/10.48550/ARXIV.2408.01361 (2024).
1.
Arora, S. et al. ESPnet-SDS: Unified Toolkit and Demo for Spoken Dialogue Systems. Preprint at https://doi.org/10.48550/ARXIV.2503.08533 (2025).
1.
Wilfong, B. et al. MFC 5.0: An exascale many-physics flow solver. Preprint at https://doi.org/10.48550/ARXIV.2503.07953 (2025).
1.
Kearns, F. L. et al. D614G reshapes allosteric networks and opening mechanisms of SARS-CoV-2 spikes. Preprint at https://doi.org/10.1101/2025.03.07.642081 (2025).
1.
Yang, Y., Taherian, H., Kalkhoriani, V. A. & Wang, D. Elevating Robust ASR By Decoupling Multi-Channel Speaker Separation and Speech Recognition. in ICASSP 2025 - 2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) (IEEE, Hyderabad, India, 2025). doi:https://doi.org/10.1109/ICASSP49660.2025.10888074.
1.
Chung, J., Zhang, C. & Chen, T. Mobility Scooter Riding Behavior Stability Analysis Based on Multimodal Contrastive Learning. in 2024 IEEE International Conference on Big Data (BigData) 6439–6445 (IEEE, Washington, DC, USA, 2024). http://doi.org/10.1109/BigData62323.2024.10825478.
1.
Hossain, R. et al. Virtual sensing-enabled digital twin framework for real-time monitoring of nuclear systems leveraging deep neural operators. npj Mater Degrad 9, 21 (2025).
1.
Pulavarthi, V., Nandal, D., Dan, S. & Pal, D. Are LLMs Ready for Practical Adoption for Assertion Generation? Preprint at https://doi.org/10.48550/ARXIV.2502.20633 (2025).
1.
Marques, J. M. C. et al. Map Space Belief Prediction for Manipulation-Enhanced Mapping. Preprint at https://doi.org/10.48550/ARXIV.2502.20606 (2025).
1.
Singer, L. P. et al. Optimal Follow-Up of Gravitational-Wave Events with the UltraViolet EXplorer (UVEX). Preprint at https://doi.org/10.48550/ARXIV.2502.17560 (2025).
1.
Tajwar, F. et al. Training a Generally Curious Agent. Preprint at https://doi.org/10.48550/ARXIV.2502.17543 (2025).
1.
Fiotto-Kaufman, J. et al. NNsight and NDIF: Democratizing Access to Open-Weight Foundation Model Internals. Preprint at https://doi.org/10.48550/arXiv.2407.14561 (2025).
1.
Bai, H., Zhou, Y., Li, L. E., Levine, S. & Kumar, A. Digi-Q: Learning Q-Value Functions for Training Device-Control Agents. Preprint at https://doi.org/10.48550/arXiv.2502.15760 (2025).
1.
Xu, Z., Gupta, A., Li, T., Bentham, O. & Srikumar, V. Beyond Perplexity: Multi-dimensional Safety Evaluation of LLM Compression. Preprint at https://doi.org/10.48550/arXiv.2407.04965 (2024).
1.
Xu, Z. et al. A Survey of Model Architectures in Information Retrieval. Preprint at https://doi.org/10.48550/ARXIV.2502.14822 (2025).
1.
Li, Z. & Luo, Y. Rewiring protein sequence and structure generative models to enhance protein stability prediction. Preprint at https://doi.org/10.1101/2025.02.13.638154 (2025).
1.
Gangwar, N., Bhat, S. P. & Kani, N. Integrating Arithmetic Learning Improves Mathematical Reasoning in Smaller Models. Preprint at https://doi.org/10.48550/ARXIV.2502.12855 (2025).
1.
Srivastava, G., Cao, S. & Wang, X. Towards Reasoning Ability of Small Language Models. Preprint at https://doi.org/10.48550/ARXIV.2502.11569 (2025).
1.
Luo, J. & Luo, Y. Learning maximally spanning representations improves protein function annotation. Preprint at https://doi.org/10.1101/2025.02.13.638156 (2025).
1.
Chen, W. et al. OWLS: Scaling Laws for Multilingual Speech Recognition and Translation Models. Preprint at https://doi.org/10.48550/ARXIV.2502.10373 (2025).
1.
Ozgulbas, D. G. et al. Probing the Role of Membrane in Neutralizing Activity of Antibodies Against Influenza Virus. Preprint at https://doi.org/10.1101/2025.02.11.637756 (2025).
1.
Abedsoltan, A. et al. Task Generalization With AutoRegressive Compositional Structure: Can Learning From $\d$ Tasks Generalize to $\d^{T}$ Tasks? Preprint at https://doi.org/10.48550/ARXIV.2502.08991 (2025).
1.
Chu, T., Estrada, J. B. & Bryngelson, S. H. Bayesian optimal design accelerates discovery of material properties from bubble dynamics. Preprint at https://doi.org/10.48550/ARXIV.2409.00011 (2024).
1.
Choi, K., Yeo, E., Chang, K., Watanabe, S. & Mortensen, D. Leveraging Allophony in Self-Supervised Speech Models for Atypical Pronunciation Assessment. Preprint at https://doi.org/10.48550/ARXIV.2502.07029 (2025).
1.
Beaglehole, D., Radhakrishnan, A., Boix-Adserà, E. & Belkin, M. Aggregate and conquer: detecting and steering LLM concepts by combining nonlinear predictors over multiple layers. Preprint at https://doi.org/10.48550/ARXIV.2502.03708 (2025).
1.
Richards, C., Dima, A., Ferguson, D. & Witek, H. Growing black-hole hair in nonminimally coupled biscalar gravity. Preprint at https://doi.org/10.48550/ARXIV.2501.14034 (2025).
1.
Osorio, J. et al. Keep it Local: Comparing Domain-Specific LLMs in Native and Machine Translated Text using Parallel Corpora on Political Conflict. in 2024 2nd International Conference on Foundation and Large Language Models (FLLM) 542–552 (IEEE, Dubai, United Arab Emirates, 2024). http://doi.org/10.1109/FLLM63129.2024.10852489.
1.
Avdiunina, P., Jamal, S., Gusev, F. & Isayev, O. All that glitters is not gold: Importance of rigorous evaluation of proteochemometric models. Preprint at https://doi.org/10.26434/chemrxiv-2025-vbmgc (2025).
1.
Pilny, A., Bonito, J. & Schecter, A. Coding Small Group Communication with AI: RNNs and Transformers with Context. Small Group Research 10464964251314196 (2025) http://doi.org/10.1177/10464964251314197.
1.
Deng, J. et al. $\texttt{dattri}$: A Library for Efficient Data Attribution. Preprint at https://doi.org/10.48550/ARXIV.2410.04555 (2024).
1.
Chen, W., Yan, B., Chen, C.-C. & Watanabe, S. Floras 50: A Massively Multilingual Multitask Benchmark for Long-Form Conversational Speech. in 2024 IEEE Spoken Language Technology Workshop (SLT) 891–898 (IEEE, Macao, 2024). http://doi.org/10.1109/SLT61566.2024.10832167.
1.
Nakamura, T. et al. Discrete Speech Unit Extraction via Independent Component Analysis. Preprint at https://doi.org/10.48550/ARXIV.2501.06562 (2025).
1.
Khot, A., Wang, X., Roy, A., Kindratenko, V. & Neubauer, M. S. Evidential Deep Learning for Uncertainty Quantification and Out-of-Distribution Detection in Jet Identification using Deep Neural Networks. Preprint at https://doi.org/10.48550/ARXIV.2501.05656 (2025).
1.
Andrews, J., Weirich, K. & Schiller, U. D. Molecular-Scale Simulation of Wetting of Actin Filaments by Protein Droplets. J. Phys. Chem. B 129, 1109–1121 (2025).
1.
Wang, S. et al. Deep CNN-based semi-supervised learning approach for identifying and segmenting corrosion in hydraulic steel and water resources infrastructure. Structural Health Monitoring 14759217241305040 (2025) http://doi.org/10.1177/14759217241305039.
1.
Feng, J. T., Satheesan, S. P., Kong, S., Donders, T. H. & Punyasena, S. W. Addressing the open world: detecting and segmenting pollen on palynological slides with deep learning. Preprint at https://doi.org/10.1101/2025.01.05.631390 (2025).
1.
Vatansever, D. & Levin, D. Collisionless Plasma Plume Expansion Under External Magnetic Fields. (2025).